Claude Codeâs Next Era
We are excited to have Anthropic share their latest AI x Finance work at AI Engineer New York, coming up in 2 weeks!
In case youâve been under a rock, hereâs a non-exhaustive list of what Anthropic has been shipping since closing the largest fundraise of all time in May at $47B ARR:
June: Launched Claude Tag and Sonnet 5 and Fable 5
Last month: Fable/Mythos 5.1, and EFS (upcoming pod)
IPO target $2T, end 2026 ARR estimated $100B
Cowork/chat merged before did
Dario endorses the same Pacing the Frontier message cosigned by all labs
Last week: Opus 5.5, Plugins portal, Cloud Sessions/Claude Projects
Today: Sonnet 5.5!
Todayâs episode should catch you up, with Thariq Shihipar, the explainer-king of Anthropic, who we last caught up on Fable launch day with The Field Guide to Fable:
The Future of Mutable Software
Pay special attention to Claude Mods (especially the cheatsheet):
In general this is also the inverse of the other viral tweet from Thariq:
Cloud Brain, Local Hands
And give a try to Claude Projects:
The âhandsâ terminology is not just an analogy for the local/cloud paradigm that is being built up at frontier coding agent companies like Cognition, but is ALSO particularly relevant to the safety systems discussions that weâll be discussing with Anthropic in an upcoming episode as they prepare to pace to frontier with responsible AI deployment.
For those who want Thariqâs writing tips we teased at the start of the pod, watch the full video here:
From the rapid rise of Claude Code to a future where agents can rewrite their own harnesses, collaborate across teams, and operate across cloud and local environments, the way we build software is changing extraordinarily fast. In this episode, Anthropicâs Thariq Shihipar joins swyx and Vibhu to unpack how power users are actually working with Claude Code today, why prompting remains a high-skill discipline, and where Anthropic thinks the agent harness is headed next.
We go deep on Claude Codeâs evolving interface: Ask User Question and elicitation, artifacts as persistent generative interfaces, Claude Tag for multiplayer agent workflows, Projects, model effort, implementation notes, and the new Claude Mods system for customizing the harness itself. Thariq explains why Claude.md may eventually disappear, why the smartest model could also become the cheapest model for many tasks, and why mutable software could become a new paradigm for how applications are built and customized.
The conversation then turns to agent security and Anthropicâs âPacing the Frontierâ argument. Thariq walks through recent incidents where agents discovered unexpected ways to communicate, exploit infrastructure, reverse-engineer benchmark scorers, and chain vulnerabilities together. We discuss sandboxing, prompt injection, autonomous agents, interpretability, constitutional classifiers, probes, fallbacks, Auto Mode, and why securing increasingly capable agents may become one of the defining engineering problems of the next few years.
We discuss:
Why agentic coding went from controversial to the default in less than a year
Why prompting is still one of the highest-leverage skills for working with Claude Code
How expert users build a mental model of Claude and what it can reliably one-shot
Why discovering your âunknown unknownsâ matters more as agents become more capable
Artifacts as persistent, generative interfaces between humans and agents
How Claude could split into a cloud-based âbrain,â local or remote âhands,â and dynamic interfaces
Claude Tag, Projects, and multiplayer agents and how collaborative agent workflows could evolve
Why spending more time on the initial prompt can dramatically reduce wasted agent work
When to use low, medium, high, or max effort for different engineering tasks
Why frontier models may eventually outperform smaller models on both intelligence and token efficiency
Why implementation notes can expose decisions the model considered but chose not to make
Why Claude.md may eventually disappear â and why starting without one can sometimes be better
Claude Mods: customizing the execution loop, UI, subagents, routing, and behavior of Claude Code
Model routers, forked agents, and supervisor agents that automatically improve agent workflows
Why Claude Mods may be an early preview of âmutable softwareâ
The bitter lesson of harness engineering and why agent architectures go out of date so quickly
How Claude Tag is becoming an organizational harness for multiplayer work
Why giving agents access to company data creates an enormous new security surface
The Exploit-Bench incident where agents discovered ways to communicate and collaborate
Why agents hacked Hugging Face for scorer code rather than benchmark answers
How agents chained sandbox and infrastructure vulnerabilities in unexpected ways
Why increasingly capable agents make traditional security assumptions harder to maintain
The argument behind Anthropicâs âPacing the Frontierâ proposal
Why software engineers are increasingly doing two jobs: engineering and keeping up with AI
Constitutional classifiers, probes, and fallbacks and what interpretability looks like in production
How Auto Mode checks whether an agentâs actions actually match the userâs permissions
Why Thariq can see serious AI risks while still having a relatively low p(doom)
Thariq Shihipar
Timestamps
00:00:00 Introduction
00:04:12 Ask User Question and the Future of Agent Interfaces
00:08:29 Artifacts, Projects, and Multiplayer Agents
00:15:37 Prompting as the Core Claude Code Skill
00:21:52 Context, Effort, and Smarter Model Usage
00:28:10 Is Claude.md Going Away?
00:32:49 Claude Mods: Customizing the Claude Code Harness
00:36:35 Model Routing and the Rise of Mutable Software
00:44:40 The Bitter Lesson of Harness Engineering
00:50:49 Claude Tag as an Organizational Harness
00:55:59 Pacing the Frontier and Autonomous Agent Security
00:58:22 Agents Hack Hugging Face for the Scorer
01:05:34 What Happens When Agents Need More Compute?
01:10:32 AI Coding Is Changing Faster Than Engineers Can Keep Up
01:17:17 Probes, Fallbacks, Interpretability, and Auto Mode
01:28:32 AI Risk, p(doom), and Closing Thoughts
Transcript
Introduction: Life at Anthropic and the Pace of Change
Swyx [00:00:00]: Weâre here in the studio with our friend Thariq from Anthropic, and I guess generally the Claude Code, I-- thereâs, thereâs so much, merging of boundaries and youâve been so on top of everything since you joined Anthropic. You have been early to Claude Code itself, but then also, and youâve told that story in other podcasts, and youâve also been talking about seeing like an agent. Most recently you did the top AIE World Tour talk, Field Guide to Fable, which obviously you guys launched Fable, so that was-- thatâs cheating. And mostly you most recently also launching Claude Tag, and weâre also gonna be talking about Pacing the Frontier. Thereâs a lot going on in Anthropic. I guess top of the question is, whatâs it like being at Anthropic when thereâs so much going on?
Thariq Shihipar [00:00:48]: I think that It is, like. I think you can get whiplash sometimes. I think, like, going. When I joined Anthropic, I joined because of Claude Code. Like Claude Code had just come out and I was like, âThis is so good.â And Opus 4 to me was like just, I could not imagine, like, how good it was? And that was, like, a real moment for me. But I was, like, trying to convince, like, my startup friends to use agentic coding, and theyâre like, âOh, no, like, our engineers donât think itâs good enough,â or something. And I was like, âThatâs insane.â and now you, like, fast-forward, 12 months, less, and, like, itâs just like, yeah, the default way that everyone codes, right? And I think that, like, just having to go from, like, selling it to, like, now, teaching people how to be. make the most use of it and be more efficient and things like that is just like a big, like big change. And, yeah, I think, like, itâs just hard to stay on top of everything as a human? Like, I think things happen so fast and like
Swyx [00:01:51]: You just throw more agents at it.
Thariq Shihipar [00:01:52]: Yeah, like thatâs like the agentic stuff scales much better than the, like, human stuff where itâs like, oh, like, there are three things happening right now and, like, theyâre all emergencies and, like, how do you, like, respond to it? Yeah.
Teaching People to Use Claude Code
Vibhu [00:02:05]: What do you split your time on? You do a lot of technical writing, engineering work.
Thariq Shihipar [00:02:10]: Yeah, so I think that, like, when I joined the Claude Code team, I wanted to teach people how to use Claude Code and I think that, like, that has been something that, like, I thought, like, maybe I would spend a little bit of time on it or, like, Iâd, like, do. I was spending some time on the agent SDK first, and I wasnât exactly sure, like, how the bitter lesson would go, when it comes to, like, harnesses, right? Like, I think sometimes we were like, âOh, like, whatâs after Claude Code?â? And so initially I was like, I just wanna teach people how to use Claude Code and make it easier to use Claude Code. And I think that has just, like, as the harnesses have gotten better and better, thatâs like the dominant problem now is, like, how do you use the agents, right? Like, itâs like such a high skill expression thing. So I do that and then I do engineering work. I give talks, but I think, like, when Iâm doing engineering work, my goal is to take that feedback that we get from users and also, like, then be able to talk about, like, hey, how to use Claude Code to do engineering. So thereâs like a good loop there. Yeah.
Swyx [00:03:07]: Yeah. Iâll-- For listeners, weâll attach, the talk that you did with Sarah for the Dev Writers, meetup
Thariq Shihipar [00:03:13]: Oh, yeah
Swyx [00:03:13]: Which we talked a little bit about, well, first you do the work and then you talk about the work.
Thariq Shihipar [00:03:16]: Right.
Swyx [00:03:16]: Something like that.
Thariq Shihipar [00:03:17]: Yeah.
Swyx [00:03:17]: Itâs sow and reap or
Thariq Shihipar [00:03:19]: Yeah, reap and. Sow and reap.
Swyx [00:03:21]: Something like that. Something like that. Yeah, so, and then just to preview a little bit, we are gonna talk about the evolution of the harness. It has come a long way from just being a CLI. Weâre gonna talk about, Claude Mods, which is starting to leak today, because you couldnât keep it secret.
Thariq Shihipar [00:03:36]: Yeah. yeah.
Swyx [00:03:39]: Yeah, thereâs, thereâs a lot, there. I think you started off with, like, adding ask user question tool, which people love and hate.
Thariq Shihipar [00:03:48]: Yeah.
Swyx [00:03:48]: Like, I thought it was, like, very innovative, and then now I have, like, my own version. You have your Interview Me version.
Thariq Shihipar [00:03:55]: Yeah.
Swyx [00:03:56]: And, yeah, everyone just has, like, their own stuff. And, like, it no longer matters âcause now youâre supposed to, write prompts that create other prompts and loops and all these things.
Ask User Question and Human-Agent Interaction
Thariq Shihipar [00:04:05]: Sure, yeah.
Swyx [00:04:06]: So whatâs the state of the art, today? Like, what are people. what are you, like, telling people to do today?
Thariq Shihipar [00:04:12]: Yeah, ask user question was the first time that the model was good at elicitation. I think this was, like, an emergent behavior that I, like, wanted to see if the models could do. I have, like a human-computer interaction background, so I, like, did that in undergrad and grad school. And so this was like. I think itâs like human-agent interaction to me, like, trying to figure out, like, how can the agent communicate with you and extract, the requirements, right? I think that, like, one of the things about, like, thatâs difficult as Claude Code has gone broader and broader is that everyone has, like, their own way of using it, and itâs very hard to, like, change the default behavior. So for example, like, if someone asks Claude Code to do something,
Thariq Shihipar [00:04:59]: Sometimes they just want them to do the work, âcause theyâre, like, maybe a very good prompter, and sometimes they want. like, are not good at prompting? And you need. like, the agent needs to, like, clarify? And so thatâs, like, a good split. Like, and the ask you the question tool like, splits along that side where, like, are-- do you feel like youâre good enough to instruct the agent as it is, or is the agent able to, like. does the agent need to, like, pull out more requirements and, like, collaborate with you more and really understand your preferences?
Thariq Shihipar [00:05:27]: I, on the whole, believe that pretty much everyone is more on the latter than the former, that they, like, have more ambiguity and they know less than they want, than they, like, think they know about the problem. but, like, itâs like a interface design problem to make that easy? And so, like, if youâre designing a problem, like, or if youâre going through a problem, like, things like whatâs the schema or, like, whatâs the call stack and things like that are really important. like, the details in the design are important. Ideally, you want to figure out some of these, like, hard problems ahead of time before starting implementation. And yeah, thatâs why they call, like, unknowns, right? And so I think that this will forever be, like, a skill in agentic coding is, like, figuring out your unknowns. So, like, because even if the model is, like, super intelligent- It, like, needs to know what you want? And, like, you have preferences. like, you need to like, pull the, pull that out. and so thatâs, like, I think how Iâm, what Iâm pushing. the question then is, like, how does the agent interact with you? And I think that has been HTML, has been, like, the big way of doing that. And weâve recently added artifacts, right? And artifacts, I think weâve done a bad job of, like, or, like, Iâve done a bad job of, like, explaining how to use them fully. We have a lot of property capabilities. They have a database associated with them? And so every artifact can store and write persistent data. They can, like, feed back into Claude? And so, like, one thing that, like, people are not doing yet that Iâm trying to, like, encourage is, like, this idea of a dashboard artifact. So you have, like, Claude working on a project long-term. Maybe itâs like a kanban or something. it can store that kanban data in its database. Multiple Claudes can access that data via, like, the artifact MCP, and, like, that artifact can, like, talk to those Claudes as well. And so, like, the. Weâre building the primitives for you to be able to have this, like, generative interface via artifacts that will, like, let you surface more of that rich detail from the agents. And I think that, like, almost everything with agents right now is, like, this problem of, like, you think what you want, but you donât really know what you want, and, like, the agents need a lot of detail, and collaborating with them in the loop is really important. And so artifacts are, like, the, like, way that weâre trying to evolve there. But thereâs a lot of work to do because itâs so much more complicated than, like, a multiple-choice question? thereâs a lot more, like, detail in terms of, like, diagrams and code snippets and schemas or, like, whatever it is for that problem. But, like, artifacts is, like, the mo-more AGI-pilled way of, like, doing ask user question. So yeah.
Artifacts as the Interface to the Harness
Swyx [00:08:15]: I think one thing thatâs unclear to me about these, the artifact stuff is, like, what feedback should go in through the artifact and what feedback should go through a Claude, a chat? Because the more AGI-pilled one is to just feed everything to the Claude.
Thariq Shihipar [00:08:29]: I think the more AGI-pilled one is to go through the artifact. Like, and I think that, like, we imagine in the limit, I think that artifacts will be your interface into the harness? You can, like, comment on this, like, live, like, document of your plan, of the work. you can see maybe, like, multiple agents and different agents are doing this, and that artifact is built for the current work that youâre doing, right? And so, like, each one has, like, slightly different. I think weâre still, like, getting there from, like, an infrastructure perspective. But yeah, I think, like, on-the-fly interface for your harness is probably where things are headed.
Vibhu [00:09:03]: Is there a version of it thatâs an abstraction from CLI or chat and you. Because right now, a lot of it is, okay, youâre interfacing with Claude Code, youâre having HTML given back for a mockup. Itâs pretty rich. Thereâs diagrams. Artifacts are ways to connect these together. Why not just do everything that way?
Separating Brain, Hands, and Surface UI
Thariq Shihipar [00:09:22]: Then it becomes, like, separating out, like, where is the inference happening? Where is the intelligence happening? Where is the work happening? like, I think this is like, difference between, like, or, like, some of the distinction between local and cloud, right? And so, I think right now, if you use Claude Code, itâs, like, local and, like, you can spin off remote control, for example, to get some cloud behavior, or you can spin off Claude Code in the cloud, right? Weâre moving towards a place where instead of Claudes, like, you message a local Claude, it starts a session locally and it executes, to more like you have a Claude that you message thatâs in the cloud thatâs running. it can run, like, local, or, like, cloud sessions. This is how Claude Tag works. But, like, over time, weâll add, like, local hands as well. And so, like, local hands will be the ability for that agent to access your computer if itâs online, and be able to, like, work there. And so it can spin off many different subagents. It can, like, commu- those subagents can communicate with each other, and thatâs where the artifact comes in to display all of that work. So you can imagine, like, the. Youâre separating out these things. So thereâs, like, the surface UI display thatâs an artifact and hosted somewhere and has a database and everything. There is the inference intelligence, right, thatâs happening on the cloud, and you donât have to worry about shutting off your computer or whatever, right? and then thereâs the, like, hands. Like, and it can be local, it can be in, like, a remote sandbox or wherever you need your work to be done. Thatâs like unpackaging, like, the Claude Code experience right now where, like, right now it all happens in one place, right? So.
Multiplayer Agents, Claude Tag, and Projects
Vibhu [00:11:00]: How do you see, like, the multiplayer side of that? So say teams want to work in this way. Right now itâs very individual, but how do you see the future of multiplayer? Like, right now, I guess thereâs Claude Tag, which is a version, but.
Thariq Shihipar [00:11:12]: Weâre launching projects. And so projects is the, like, this abstraction thatâs like Claude Tag, but on our Claude products, right? So you can message it and, like, it will do the Claude Tag-like stuff, like spinning off subagents. So We think with multiplayer. Like, Claude Tag is, like, a little bit more native multiplayer because itâs just, like, in your Slack and the permissions are all figured out and stuff like that. But I do think multiplayer is, like, an important part of the story and, like, that will need to get tied together more. Like, you can imagine how complicated it gets when youâre like, oh, you have hands, but now you have other hands in other peopleâs computers too, and, like, you need to, like, permission them or, like, you have, like, your MCP and someone elseâs MCP, and how do you figure out how to use them, right? It gets, like, quite complicated. And Claude Tag does a good job of, like, sanding down all of these issues, right? So that, like, when you have, yeah, Google Docs, how does it access Google Docs, right? Like, it accesses through the shared Claude MCP, or it can access through your local credentials as well if it doesnât have access. But yeah, I think Claude Tag is our multiplayer, product, and itâs really useful for these, like, things that are inherently multiplayer. Like, okay, like on-call, for example, incidents are inherently multiplayer. You want to tag Claude, you want multiple people to log in, you want it to be able to find context. I think whenever Iâm, like, working on something and I want, like, privacy or security or, like, I want other people to review itâs really nice to, like. Iâll have a channel per project and Iâll, like, at legal, for example, be like, âHey, like, I want to ship this. Can you, like.â Like, hereâs. Like Claude knows everything, just chat with it. And that way legal gets precise answers, on like what exactly is shipping into the code, and I donât need to be in the loop, right? So I think like multiplayer is getting like more and more like, yeah, everyone can participate with Claude. I think Claude Tag is like that product and like projects will start off single player and will like, expand.
Swyx [00:13:14]: I think thereâs a question about like maybe dual questions about identity and the unit of isolation.
Identity, Permissions, and Isolation
Thariq Shihipar [00:13:20]: Yeah.
Swyx [00:13:20]: Claude Tag, you specifically chose to make it its own identity
Thariq Shihipar [00:13:26]: Yes.
Swyx [00:13:26]: Which is like, a controversial choice. Thereâs, thereâs other ways to do it.
Thariq Shihipar [00:13:30]: Yeah.
Swyx [00:13:30]: Claude Projects probably it sounds like, if itâs anything like ChatGPT Projects, it is, the isolation is that artifacts, that cloud instance, everyoneâs collaborating on this. Itâll. It sounds like, it should be like if youâre, if youâre collaborating with legal on a thing, like that channel should be a project, right? Like itâs not yet
Thariq Shihipar [00:13:50]: Yes.
Swyx [00:13:50]: But it. thatâs the natural next step.
Thariq Shihipar [00:13:53]: Yeah, like I think in Claude Tag, itâs effectively. Like Claude Tag, you have to do your own arrangement. And so Claude Tag, yeah, each channel is like you can name it as you want, and I name
Swyx [00:14:04]: Yeah.
Thariq Shihipar [00:14:04]: Like each feature
Swyx [00:14:06]: Yeah.
Thariq Shihipar [00:14:07]: As a channel.
Swyx [00:14:07]: And, but I think like there is some trans- like itâs unclear when there is transference, because letâs say it is. if you have a coworker
Thariq Shihipar [00:14:14]: Yeah.
Swyx [00:14:14]: Who is tagging on all these things, yes, there is transfer
Thariq Shihipar [00:14:16]: Yeah.
Swyx [00:14:16]: Because itâs the same person. but with Claude, itâs unclear if itâs like necessarily like, well, no, you donât know any of. you donât know about the other stuff. You should only use this stuff.
Thariq Shihipar [00:14:25]: Itâs like the tip of the iceberg meme, right, where you can like. This is what we spend so much time on
Swyx [00:14:31]: Yeah.
Thariq Shihipar [00:14:31]: Is like there is like infinite surface area of like, okay, you want Claudes to. Not infinite, but like thereâs like surface area, a lot of like, surface area to figure out of like permissions and visibility and like how can you let Claude operate as well as you can, as safely as you can? And obviously, this is very important to us because like security for our code base is very important. And so weâve put a lot of time into this. Yeah, thereâs so many like edge cases you can figure out where itâs like, oh, like, yeah, this Claude in this channel has different permissions, but it can message another channel, and canât it exfiltrate data that way? Or like can you like. What if it uses your MCP and then messages someone else? Like thereâs like so much, and weâve like really put a lot of work into sanding it down.
Swyx [00:15:14]: Yeah. Lots of work. okay. Fable?
Fable and the Meta-Skill of Prompting
Vibhu [00:15:18]: Fable, you wrote two good articles. youâve written many good articles
Thariq Shihipar [00:15:22]: Yeah.
Vibhu [00:15:22]: But on, Field Guide to Fable, Building Claude Code. Iâm curious from what youâve seen, is there any common patterns that you see in like top users at Anthropic externally? Like what are best practices for getting the most out of Claude Code?
Thariq Shihipar [00:15:37]: The like meta skill I say is like prompting is like very important? And like that. Like I think this is like not trivial to say because I think a lot of people are like, âOh, prompting doesnât matter. Itâs just like I can just say a sentence and Claude will do it.â And I think prompting is really this like, this. Itâs like public speaking, like, or writing or something, and for a specific audience, and that audience is Claude. And you need to like build a mental model of Claude and how it thinks and how it works, right? And so thatâs like the most important skill in working with Claude Code is like having this mental model, right, of Claude and like what it can do well, what it can one-shot, what it canât. And so many people when you see prompting, theyâre just like, theyâre short prompts, but they have such a good mental model of Claude and of like the code base and things like that like itâs effortless? But itâs like high skill ceiling. So like that work of like, spending a lot of time prompting and building mental models of how, and intuition for how the agents work is really important. And then I think like the next thing is like the unknown stuff we talked about earlier, where itâs like being able to find out like your, what you donât know or what you havenât written down, learning about like different things. I think as Claude can do more and more things, the likelihood of you doing something out of distribution for you and like you have low domain knowledge on is very high? And the more you can like learn the vocabulary to be able to prompt Claude, it becomes really important. And so like I think the most important unknowns are the unknown unknowns, where youâre like, I just like donât even know that this exists, right? Yeah, exactly. I think thatâs like a illustration of like the map and the territory, right, where youâre like, âOkay, this is my prompt,â and the territory is like the actual like work that the agent needs to do, right? And if you are like very precise, you can give more precise things, right? So like for example, in design, Iâm not very precise. Iâm not a designer, so I say like, âGive me like eight different mock-ups.â But if I was a designer, maybe Iâd be like, âOh, hey, here are some reference sites.â Like, âI want this type of font and this type of like look to it, and hereâs like a few different components to like visualize. Hereâs a Figma MC board to bring in,â like. And so you can just be so much more precise with that language. And if youâre not a designer, you just need to like try and learn the language or learn the unknown unknowns. And this is true of like everything, I think. Like the more, like you can work with Claude to learn like how things work, the better your prompting will be. I think another good example of this is like game design, like where a lot of people are like, âOh, like I can vibe code a game now.â And theyâre like, âItâs not fun.â And like itâs just like the thing about game design is like every one of these choices has like a lot of
Taste, Domain Knowledge, and Learning the Vocabulary
Swyx [00:18:25]: Variations.
Thariq Shihipar [00:18:25]: A lot of like craft to them. So itâs like, oh, okay, like when youâre making a flying game, the feel of the plane and the like, way it responds to your controls has a lot of like. Like, a game designer would spend like days on that. Do? and like
Swyx [00:18:44]: To me, thatâs what taste is, right?
Swyx [00:18:45]: Like it is like from the possible space of one thousand mathematically valid answers
Thariq Shihipar [00:18:49]: Yeah.
Swyx [00:18:49]: Hereâs the one that is the humans will like.
Thariq Shihipar [00:18:51]: Yes. Yeah.
Thariq Shihipar [00:18:52]: I think with taste, Iâm like torn on this word âcause I think youâre right, but everyone has different definitions, and it sounds kind, sounds like low skill or like elitist almost, where youâre like, oh, like there are certain people with taste?
Swyx [00:19:06]: Itâs like taste is what I call taste.
Thariq Shihipar [00:19:07]: Yeah, exactly.
Swyx [00:19:08]: And itâs like these guys donât have taste.
Thariq Shihipar [00:19:09]: Yeah, exactly. Oh, like an engineer doesnât have taste. Like I, the like founder, have taste.
Thariq Shihipar [00:19:14]: ? And I think thatâs not true. Like I think like the engineers have a lot of taste for these particular like problems? And I think everyone has taste for particular problems. I think like Jason Liu, like say like in order to, yeah, have taste, you have to eat?
Thariq Shihipar [00:19:32]: And I really like that, where itâs like, okay, you have to like do a lot of things. You have to like iterate and figure out what you want, what you like, and, like build that like domain
Swyx [00:19:41]: Yes
Thariq Shihipar [00:19:41]: Domain vocabulary. And then when youâre prompting, youâre like synthesizing all of that for a product.
Swyx [00:19:46]: Isnât it annoying when someone else says it better than you?
Swyx [00:19:48]: Itâs just like, fuck, I have to quote this guy forever.
Vibhu [00:19:51]: Having to quote Jason Liu forever.
Vibhu [00:19:53]: Heâs gonna love this.
Thariq Shihipar [00:19:55]: So I get prompts, more than that.
Vibhu [00:19:57]: And sometimes itâs not even that. Sometimes itâs just intuitive, right? Like you donât realize you even want something till a model puts it out, and youâre like, âOh, this just feels immediately better,â right?
Voice Prompting and Information Density
Thariq Shihipar [00:20:07]: Yeah, exactly.
Swyx [00:20:09]: One thing I go back and forth on is I feel like the way I prompt half the time, letâs say I use voice.
Swyx [00:20:16]: Did I say voice? Other people have voice. that is the opposite. That is just like me rambling for like two minutes Pressing down the function key and then let go, and then like hopefully it figures it out. And oftentimes it does.
Thariq Shihipar [00:20:26]: Yeah.
Swyx [00:20:26]: But itâs not as thoughtful as like a structured prompt with like Well-run communication as though itâs a PRD or a memo. Is that in line with how people do this? Thereâs like bimodal prompting where thereâs some prompts where you spend a lot of time upfront and other prompts you just dash it off?
Thariq Shihipar [00:20:43]: I donât think the voice is necessarily low. Like I think itâs like more like how much information is in the prompt. like the model can. Like you can and like add some sentences
Swyx [00:20:53]: Right
Thariq Shihipar [00:20:53]: And be like, âOh, like I changed my mind,â like in the middle of the prompt, and it will be able to follow that perfectly? So I think the like actual format of the text is less important, but then like the ability to. Like how much information is in it, right? And I think for voice, a lot of times, going back to like human-agent interaction and like for a lot of people, itâs just way easier to talk than to like type? and I. If that gets more information out of you, like thatâs better.
Vibhu [00:21:21]: At some level, it feels like just giving the model as much context
Thariq Shihipar [00:21:24]: Yes
Vibhu [00:21:24]: Over prompting before you kick off is a best practice. I donât know. A lot of the times, like when I was first trying out Fable, I spend a solid 30 minutes like really crafting a long prompt. This, I think, is a response of models running for longer and longer, right? Itâs still a little difficult to nudge them as theyâre in like, in the loop, but I just like intuitively spend more time kicking off that first prompt and working with it a lot.
Spend More Upfront, Iterate Less
Thariq Shihipar [00:21:52]: My personal opinion is that if I was a software engineer, if I was like, just running my own startup, for example, I think I would mostly fit, stick to a max 20x? like maybe verification and so code review are like separate things. But I think like what I see a lot of times is people hit rate limits when theyâre doing this like, oh, like it did a lot of work and youâre like, âOh, I donât like this.â Like, âCan you like undo this and redo it?â And then youâre like iterating on this like thing that the model could have done if you had like spent more upfront time or given it better context? And instead itâs like youâre like, âNope, donât like that design. Try this.â Or like, âYou messed this up,â or something like that. And then that just eats up so much more of like, your usage. And so thatâs like, I think maybe like a key like tip both for like efficiency as well, right? And yeah, I think like context, and not just like context on like what the goal is good, right? Like are you building a prototype or is it like a production thing? Like where can you spend compute or when, where can you not spend compute? Like I think you have to give the model permission or like not permission to do things sometimes where, like it doesnât know intuitively how much you want to spend on this task, right? And you can use effort for this. So I did-- Iâm working on a blog post about that where itâs like, if you want. For like we see that effort scales with the complexity of the task. So for security, effort gets like way more results. Like high effort versus like low effort gets, like changes the evals a lot. But for software engineering, it doesnât change it a huge amount because effort is mostly spent on the verification and the like edge case testing and things like that. And so like being able to like give the model that guidance of like, âHey, this problem is something that I think I want you to spend a lot of time verifying and edge case testing,â?
Effort, Model Choice, and Verification
Vibhu [00:23:43]: How about model in the mix? So, thereâs Opus and Fable with effort.
Thariq Shihipar [00:23:47]: Yeah.
Vibhu [00:23:48]: Thereâs also Haiku in there.
Thariq Shihipar [00:23:49]: Yeah. Itâs not quite true yet, but itâs very close where I think the frontier models will be Pareto dominant over like almost everything. like maybe. And sometimes I think Opus might be Pareto dominant. Do? Like I think depending on like how things, like shake out if itâs like a newer version of Opus. But I think that like increasingly itâs just going to be like the smart model is going to be able to like do the simple task for less tokens than the like the other models because of verification. With verification, in the limit, your model doesnât need to verify, right? If itâs a perfect model, it just does the work once and itâs like, okay, like you, I did it? And increasingly with Fable, Iâm like, Iâm like, âDude, you donât need to spin up Chromium and screenshot all of these things.â Like I see it. Like you did it, right? And so a lot of the. At higher effort, you spend more of those tokens verifying. But if youâre working on simpler problems, and a lot of software engineering is like well, like in Fable, like low and medium stability, it can spend less tokens verifying. And as the models get smarter and smarter, they will just be able to like, âAll right, done.â? Like, I can run the lint for sanityâs sake, but, like, I, like, know it lints? Like, you donât even need to do that. And that will be so much more token efficient than, like, the smaller models. Yeah.
Swyx [00:25:15]: Is there a good, practice on our side that we can use to see if weâre using too much effort? Like, I freaking
Thariq Shihipar [00:25:23]: Yeah
Swyx [00:25:23]: Hate wasting time on that stuff.
Thariq Shihipar [00:25:24]: Yeah. I know what you mean. I think, like, so in this blog post, my rough distribution is, like, code review and security should be, like, high or max and, like, software engineering
Swyx [00:25:37]: You said recommend mix settings per domain.
Thariq Shihipar [00:25:37]: Yeah. I think, like, if youâre doing, like, UI or something like that, like low and medium, I think is youâre building, like, an API and you want to make sure, like, you cover enough edge cases? And so I think building, like I said, that mental model of, like, how things work across these distributions is, like, yeah, part of the job.
Implementation Notes and Decision Logs
Vibhu [00:25:56]: This is more intuition-driven or eval? Because Iâm guessing this would change as you go.
Swyx [00:26:00]: He has evals.
Thariq Shihipar [00:26:01]: Yeah. So what I did in the blog post is I go over all of the terminal bench evals. So there are, like, 70 problems and Iâm show that, like, okay, like, in the security problems it does more. and then I also, like, look at some of the transcripts just in terms of, like, how-- what does it answer, what does it forget or something. And a lot of times, this is another prompting tip I have, is, like, asking it to make decision notes or implementation notes because, in every eval problem that it faces, it thinks about the correct solution, and decides not to do it. itâs like, oh, like, here is the answer. What if I did this? And then itâs like, oh, probably not? and then keeps going. And this is, like, the majority of the failures, at, like, a higher max level. Itâs very rare that the model just doesnât know how to do something. If you just have these implementation notes, then you can review and you can be like, âOh, I want you to do this thing that you didnât do.â The models are getting better at surfacing that overall. Like, I see in the transcripts of Fable 5.1, like, when it does this output, it will call out its decision-making as well. but making this more explicit in the harness is better. And now weâre, allowing ways of you modifying the harness so you can, like, add some
Vibhu [00:27:23]: Ooh.
Thariq Shihipar [00:27:24]: Calculate with there. Yeah.
Swyx [00:27:25]: Yeah. So I do wanna call out two things that you mentioned that I think exist outside of prompting. One is like, letâs, letâs call it the prompt that is so important that it shouldnât be in a prompt. It is in Claude.md or Agents.md
Thariq Shihipar [00:27:38]: Yeah
Swyx [00:27:38]: Which is like goals, right? Like your situation, your goals, the things that you want, the thing. and then second of all is the decision log or the experiment log or whatever log of traces that you might want to survive the current session to do those things. Those are, like, externalities that thereâs no standard. Thereâs no-- Itâs not like skills. Itâs not like MCP. Thereâs no standard. Itâs, itâs just like itâs a markdown file. first of all, is that right? Is Claude.md going away? You have a documented dislike of, Agents.md, but youâre gonna do it?
Claude.md, Agents.md, and Model-Specific Instructions
Thariq Shihipar [00:28:10]: Yeah. Okay. So Agents.md, yeah, like, weâre, weâre gonna do it. I think itâs just, like, different models are very different from each other? But I realize that itâs, like, such a pain to, like, maintain different ones? And yeah, like, as the models get better and better, the floor of how they accomplish the simpler task is better. And so I do think in the limit, Claude.md goes away, and maybe not even, like, that far. Like, I think, like, I think that right now it might be better to start a new project without a Claude.md.
Swyx [00:28:44]: Yes.
Thariq Shihipar [00:28:44]: I think that, like, maybe if you see very repeated failure modes, you add them to your Claude.md. The really tough thing is that this changes per model. And so, like, if youâve added a bunch of failure modes or, like even
Swyx [00:28:57]: So you need Fable MD, you need Opus MD.
Thariq Shihipar [00:28:59]: Or well, even Fable 5.1 versus Fable 5.
Swyx [00:29:03]: Yeah.
Thariq Shihipar [00:29:03]: Like, it is annoying. Like, Iâm not like,
Swyx [00:29:05]: Yeah
Thariq Shihipar [00:29:05]: Like, we donât, like, do this on purpose? Itâs just, like, how the models work, right? And so, like, maybe, like, Fable 5 had this, like, failure mode that Fable 5.1 doesnât. And if you keep this context, this running log of a bunch of different failure modes, they will probably over constrain Claude? And so this is like. we just added evals plugins for skills.
Swyx [00:29:28]: Yeah.
Thariq Shihipar [00:29:29]: And so now you can eval if a skill is better. I think Daisy on our team did this. And so, yeah, this is like weâre trying to work on this. We know itâs, like, you still have to spend tokens on it and, like, itâs not, itâs not perfect, but itâs, like, weâre trying to help out with this problem.
Swyx [00:29:44]: And so, and as far as prompting goes, the one tip I wanna offer is, something I have told people a lot is sufficiently advanced prompting is indistinguishable from sufficiently advanced executive communication. So Iâve referred to-- This is an executive comms workshop from Heavybit that is the best Iâve ever seen in my career. And they teach this thing called the SCQA model. Just Google it. Itâs a, itâs a thing. Like, people have done prompting for decades. Itâs just called executive communication. Itâs like when one person has to communicate to thousands of people down the org chart, this is what you do. so situation, complication, question and answer, is how you write the memo. but obviously sometimes you donât have the answer, but you can at least list out the SC and Q, and then they have some examples in there. So just leaving breadcrumbs for people if they want to explore.
Underrated Prompting Patterns and ELI5
Vibhu [00:30:31]: Before we move on, I wanna ask you, any other underrated tips, ways people could get a lot of value from Claude Code that theyâre not using?
Thariq Shihipar [00:30:41]: Yeah, I think a lot of them are in the, this unknowns, like, doc. Like, I give a bunch of example prompts, like, using it for brainstorming, using it to quiz you after. we added this, like, explain it like Iâm five skill which is a very short prompt. And it doesnât even say explain it like Iâm five. Itâs like the key word of this prompt is big pictures, few words. like, thatâs like the main thing. And it is shockingly good? Like, you, like, I think I tweeted about this and itâs like /eli5, and, like, you can install it as a plug-in. But yeah, itâs, like, way better at just cutting through the BS and being like, yeah, exactly right here. So the diagrams are, like, quite clear. I think one of the things that is true with artifacts is, like, they put too much text in and people are not reading the artifacts? And so, like, this simplifies it a lot more. And, yeah, this came out of, like, just people at Anthropic, like, going through very complicated incidents and being like, âWhat is happening?â? So, this one I think is great, yeah.
Swyx [00:31:47]: My version of this is the, itâs like test your understanding. Give you a few choices and then, like, if you get it wrong, you have a mismatch between what you think is happening versus whatâs happening.
Thariq Shihipar [00:31:58]: Yeah. I think this is one of those things that everyone loves talking about, and then very few people really do. Like, I think
Swyx [00:32:05]: Really helpful.
Thariq Shihipar [00:32:07]: Yeah. But most people just donât want to get quizzed about something? Unfortunately, I think this is one of the, like, things that we need to, like.
Swyx [00:32:16]: Whatâs the opposite of ask you the question or ask you the question before the thing?
Thariq Shihipar [00:32:19]: Yeah.
Swyx [00:32:19]: This is after the thing.
Thariq Shihipar [00:32:20]: Exactly. Yeah.
Vibhu [00:32:21]: Itâs a good way to stay grounded of, like, do you even know what youâre doing, right? The worst case is when people send you slop and they havenât understood what theyâre asking for or what the output is, and itâs like, âDude, I donât wanna read this. Do you even know what it is?â So, you make it a rule for yourself that before you send stuff, you should at least know whatâs implemented.
Claude Mods: Customizing the Harness
Thariq Shihipar [00:32:41]: Yes, but so you could make this a mod and you could build your own mod to, like, make sure you test it. So yeah, you can do that.
Swyx [00:32:49]: All right. Letâs get right into it. What is Claude Mod, and what is this diagram showing?
Thariq Shihipar [00:32:54]: Yeah. Okay, so Claude Mods is you can customize the entire Claude Code harness, and weâre going to. If you have requests, we will, like, let you, like, please let us know. Weâll add more and more. This works for CLI, it works for desktop. maybe it will work for Claude Tag in the future. I donât know. Like, weâre trying to make this very extensible. You can see this reference sheet. I donât want people to get overwhelmed by it? At a high level, you can customize both the execution of the harness, and the UI of the harness. And so, like, you say on that Tetris example from Boris, thatâs like customizing the UI, right? Like showing, like, Tetris in the game.
Thariq Shihipar [00:33:35]: But, like, letâs say that you wanted to do this thing where you had. you tested your assumptions or, like, tested your understanding after every project, right? What you would do is you would ask Claude to make this plug-in. It would spin a classifier after every prompt. And so, like, at the end of each turn, you would spin off a sub-agent or, like, a forked agent. A forked agent is, like, maintains the prompt cache, right? So itâs like a, like one of those unintuitive things where you can fork and do, like, a little request, and itâll be very cheap because the entire prompt cache is, like, done. And so you can be like, âHas this task been completed?â like
Swyx [00:34:18]: This is how you do BTW and all those.
Thariq Shihipar [00:34:20]: Yeah. The underlying forked agent, yes. But so you can, in the f-fork sub-agent, you can say, like, âHas this task been completed? If so, return true.â And then in your hook, or in your, like, plug-in mod, or sorry, like, in the sub-agent probably, you would say, like, âIf true, give me a quiz.â give me questions and answers, and then, like, in a JSON format, and then youâd parse it, and then you display above the prompt input, this list of questions, right? And so this is something thatâs, like, slightly token-intensive because, like, you have to do it after every end of the assistant turn. But itâs, like, a lightweight classification, and then you can, like, get this quiz, and then youâll see, like, Claude will always do it for you. You donât need to remember to do it. There are lots of these, like, tips that weâve talked about, right, where itâs like, oh, implementation notes. You can also add a tool for implementation notes now. And so, like, this tool that Iâm adding is, like, register, like, I think assumption is what Iâm calling it, but, like, maybe Iâll change it around. And this is a mod. And so, like, you give it a register assumption tool, and then it will keep a list. Itâll. Every time it does itâll keep a, like, add to the list, and then at the end it will display those assumptions? Another mod Iâm working on is a model router. And so, like, internal, like, Claude model routing, right? So itâs. This is, I want to say the reason we donât do model routing by default is, like, itâs a hard problem? And like
Forked Agents, Assumption Tracking, and Model Routing
Swyx [00:35:51]: You will get it wrong.
Thariq Shihipar [00:35:52]: Yeah, you, like, yeah, you will, like, accidentally use, like, Fable for a hard problem or Sonnet for
Swyx [00:35:57]: Yeah, if you have auto approve, but you donât have auto mode.
Thariq Shihipar [00:36:01]: Well, you will have auto. Like, you donât have, like, auto routing or something.
Vibhu [00:36:04]: You donât have auto mode for model picker.
Thariq Shihipar [00:36:06]: Yeah, exactly. So
Vibhu [00:36:07]: Iâm getting the rough question of, like, how much do you open this up and how much do people have to think about this? Like, when you talk about prompt caching and building a router, it seems like you could easily build a mod that routes per query, and Iâm just killing my plan very fast, right? I guess my question is more so, like, what is, like, a product talk like this look like, right? Who is it for? Is it for power users? Is it everyone should be able to go through
Swyx [00:36:33]: Oh, definitely power users, right?
Thariq Shihipar [00:36:35]: Yeah, I think it is power users, but, like, the nature of Claude Code is that so many people are power users? Because itâs easy to share things, like you can. Like, one person can make a good model router thing that doesnât break prompt cache all the time, and then you can, like, compose them. Another cool thing about the plug-ins is that they can hook into and compose with each other. And so I have, like, a mod that will, like, create a mode selector at the top, and any plug-ins can register to be a mode. And so, like, the auto router can be a mode, right? Or, like, you can have a mode thatâs, like, artifact mode, where itâs like it primarily talks to you in artifacts. like, you can toggle between plan mode? And so, like, you can create more and more of these modes. But the ability to create modes is in it itself a mod? And so thereâs a lot of richness here, but we do want to make it fairly easy. We want to be-- make it so that you can just, like, install someone elseâs. You can ta-- you can chat with Claude and, weâll, like, make sure that it understands the nuances of things like prompt caching and stuff, so it can, like, warn you. This is, like, not extremely complicated behavior for Claude, I think, but we should have just a good skill on how to make mods. and yeah, weâll see how we go. But I do think that this is, like, a preview of, like, mutable software, and, like, how, like, generative software, just like you can customize safely. If enabled, you could customize any piece of software. And I think that more and more apps ideally do something like this?
Power Users, Modes, and Mutable Software
Swyx [00:38:13]: And by the way, you, we have, you have another cool tweet about how, thereâs the infinite money button, which is like make your SaaS, consumable by agents. I think mutable software is interesting and, other people have also tried to do it. I think the hurdle comes when you can do everything, then people, users get, tend to get confused. So usually the stuff that works is just like one opinionated flow. This is in the side of less opinionation. Itâs just like, well, more power to power users. And I think probably unlocked by AI, where, like, you can just prompt for whatever the thing is.
Thariq Shihipar [00:38:47]: Yeah, or there can be a skill that gives the opinions?
Mods vs. Hooks vs. Artifacts
Swyx [00:38:50]: Yeah.
Thariq Shihipar [00:38:50]: And then, yeah.
Swyx [00:38:51]: So knowing a little bit about, like, TypeScript and build systems and all these things, the closest-- Iâm very curious that the team who worked on this, if, I donât know how close you were to them, if they drew any inspiration from build systems like Babel, Webpack, all these, like, old school things. Because it sounds very similar, like the plug-in ecosystem of those things where they can compose with each other.
Thariq Shihipar [00:39:11]: Yeah, Iâm not deep in the technical details, but I do know it was a collaboration with someone on the Bun team and someone on the Claude Code team.
Swyx [00:39:17]: Yeah, itâs a build system mecca.
Thariq Shihipar [00:39:19]: Yeah. Exactly. Itâs, itâs very exciting. But yeah, like, agents can just do this very complicated like, extensibility into your software now. And so, yeah, like, another reason to, like. If you run a startup, like, you can just prompt Claude and be like, âHey, like, could we make an extension system? Like, what would that look like?â?
Swyx [00:39:37]: Yeah.
Swyx [00:39:38]: And I just really wonder, like, you had hooks in the past and plug-ins, all these things. So what specifically will mods be able to do that those things could not do?
Thariq Shihipar [00:39:47]: Internally, we were originally calling this function hooks. And so, like, thatâs, like, gives you a little bit of an idea where, like, hooks register a, like an event to happen and then, like, a script to call. And this inside of the, like, TypeScript runtime is running things. And so, like, you get some benefits of just, like, it has a bunch of things in the Scope with, like, for example, like how many turns is in this conversation, right? Like, how many tokens have been used? Like, et cetera. Like, what are the messages? Things like that. So it has a bunch of messages that can be used. And then itâs just, like, a lot more hooks. So we have, like, or a lot of, lot more, like, things you can register on. And then you can do because of the. because itâs all happening in process, you can, spawn sub-agents, with four contests and contexts and stuff. And, like, that will return. You can parse the results of those. You can use structured output to like, return them. and then you can modify the UI, which you can never do in hooks. So, yeah.
Swyx [00:40:50]: Yeah. Yeah. So modify UI, this is why you showed the Tetris example. Does it also ex-extend to artifacts? I assume it does.
Thariq Shihipar [00:40:57]: You-- Like, artifacts are like a different way of customizing it. like, you can definitely. One of the mods Iâm working on is, like, this dashboard mod, which will, like, prompt Claude to maintain a dashboard, thatâs an artifact. But theyâre like, slightly orthogonal, or not orthogonal. They compose with each other in different ways. Like, mods are, like, a little bit more, like, in your Claude Code harness, changing the agent loop? And, like, the UI is, like, an added benefit. and then artifacts are just like you want to, see things at a high level, very inter- highly interactive. like, the affordances can be a lot bigger than, like a TUI or even in our desktop.
Next Steps, Supervisors, and Persistent Guidance
Vibhu [00:41:40]: Iâm guessing youâll have a good blog post on the differences, because right now you can also, make a loop that outputs to an artifact thatâs an interactive dashboard, but you can also do it with a mod. Thereâs just some thinking about making a hacking on a harness when we donât know much about the harness, right?
Thariq Shihipar [00:42:00]: Well, something Iâm excited about with mods is, like, thereâs so much things with Claude Code that you just have to remember? Youâre like, âOh, like, let me do this, and then let me call the dashboard skill that does the loop,â and things like that. And, or like, âLet me test my assumptions afterwards.â And I think, like, if you do all of these things using these little classifiers and stuff, and youâre like, âThese are the things I care about. This is what I want to do,â you can, like. You donât have to remember as much. One more, like, mod Iâm working on is a next steps mod that
Swyx [00:42:28]: I have-- I was gonna say, I have a next step skill. I always run next steps.
Thariq Shihipar [00:42:32]: And does it have access to your skills? Like, this is one of those things where Iâm like.
Swyx [00:42:37]: I think so.
Thariq Shihipar [00:42:38]: Okay. Yeah, probably
Vibhu [00:42:39]: Do skills need specific access to
Thariq Shihipar [00:42:41]: Well, I think thereâs
Swyx [00:42:41]: Donât they always have
Thariq Shihipar [00:42:42]: I think thereâs, like, specific prompting, I guess, to, like, know your skills. Like I think Claude forgets them sometimes throughout, like, the thing. But anyways, the idea of, like, yeah, next steps that also are like, âOh, hey, this has happened. Use the explain skill to explain to you what happened because this seems, like, quite complex,â? Or, like, yeah, âUse your unknown skill. It looks like you are, like, asking the model to, like, iterate on these small changes. It seems like you could prompt better.â like, âWhat if you did this?â Right? So, I think, yeah, like spending more compute there. Yeah.
Swyx [00:43:20]: And it should always come out as multiple choice. we have, I have
Vibhu [00:43:23]: We have his skill.
Swyx [00:43:24]: My next step skill is like this.
Thariq Shihipar [00:43:26]: Okay, perfect. Yeah.
Swyx [00:43:27]: You can steal it.
Thariq Shihipar [00:43:28]: Yeah.
Swyx [00:43:29]: Like, but like, for me, itâs all-- I think models really always need to be reminded, what are you trying to do here?
Thariq Shihipar [00:43:35]: Yeah.
Swyx [00:43:35]: Look at the whole transcript and go like, oh, was this original goal? Did your solution solve it? Were you lazy? If youâre lazy, maybe thereâs a reason. Maybe you needed approval from me. Maybe you needed, thereâs two things you wanna suggest. So itâs, itâs a little bit like the modification of the ask user question or interview me skill. so itâs next steps.
Thariq Shihipar [00:43:55]: Yeah, exactly. And again, the benefit of doing it with mods is you can do it as a fork sub-agent, and so it doesnât remain in the context afterwards. So you have this, like, idea of like, okay, the model is doing its execution and you have this almost like supervisor, like, that is like making sure that you can do like the next steps well. So yeah.
Swyx [00:44:15]: Yes. I do have two panels and like I often try to have a supervisor thing, keep the high-level context and then the implementation
Thariq Shihipar [00:44:21]: Yeah
Swyx [00:44:22]: Detail in another agent.
Vibhu [00:44:23]: I feel like a lot of this abstracts away as models change? The, like, half an hour ago you said bitter lesson of harness engineering
The Bitter Lesson of Harness Engineering
Thariq Shihipar [00:44:31]: Yeah
Vibhu [00:44:31]: And weâre on the other extreme right now, I feel.
Swyx [00:44:33]: Well, so yeah, exactly. If everythingâs customizable, what is Claude Code, right?
Thariq Shihipar [00:44:37]: Yeah.
Swyx [00:44:37]: And which I talked to you about last night.
Thariq Shihipar [00:44:40]: Yeah, I think that this is. I think the bitter lesson is unintuitive? In terms of like. Also, like weâre misusing a little bit of the bitter lesson here where itâs like, itâs more about like scaling and compute and stuff. But like, I think there is something where itâs just like. I think I use it as an approximation here to say that harnesses go out of date very quickly? And like how, but how they change is unintuitive? And so like the big obvious example is like from chat to like agents where you had to give them entirely new tools, right? But like, I think this new version of like, oh, it can modify its own harness, right? This is like, an own harness loop is like a way of using its capabilities, right? Or like it can build an artifact. And like, I think the way I think about it is like the models have more and more intelligence, and theyâre like so much more intelligent now than like the average software engineering task. Like, you look at the like terminal bench ones and theyâre like solve like the Jacobian conjecture. Not really, but like, itâs like theyâre, theyâre quite complex. Like, I would not have been able to do this really as a software engineer.
Swyx [00:45:42]: And you said TB4 or TB2?
Thariq Shihipar [00:45:43]: TB3. TB3.
Swyx [00:45:44]: TB3.
Thariq Shihipar [00:45:44]: Yeah. Theyâre quite complex, but the goal is still to deliver user value, right? And like you said, thereâs like this infinite space of things to do. And so the ways like you spend compute are to keep the user in the loop and make sure that like youâre getting to the right decision in the end of the day and like the right output. And artifacts and mods are this way of like spending that intelligence. and I think thatâs like, yeah, the next step. And so, yeah, I think Claude Code is like, has the core things of agent loop which are, have gotten more complicated. Itâs like, it needs a sandbox to operate safely. It needs auto mode to like make sure like the permissions
Vibhu [00:46:21]: Approvals.
Thariq Shihipar [00:46:21]: Yeah, approvals. it needs computer use and MCPs and like all of these like ways of accessing your data, and it needs web search and web fetch. And like, so the-- as the models can do more and more, the core harness has to be like quite complex and very secure. But then like how you interact with it can change quite a lot.
Vibhu [00:46:42]: What other harness engineering best practices have you, from the Claude Code team itself? I feel like, there was a phase of plan mode, which is not as used. We now have auto mode. at a point you cut the majority of the system prompt, you got rid of examples. What other best practices are there for harness engineering?
Core Harness Primitives and Managed Agents
Thariq Shihipar [00:47:02]: I think there is like a forking path where at some point, eventually, yes, the model will just be able to like vibe code the exact version of Claude Code, even describing all this complexity that Iâve talked about, right? Like auto mode and computer use and stuff. Eventually, the models will just be able to do that in one shot. But I think they can one shot simpler harnesses? And so like, I think some people. Sometimes you donât need this full, like if you donât need computer use or like all this like more complicated stuff. I think before we, you had to use things like the agent SDK, which was like Claude Code wrapped, in order to like. And I would, like suggest people do that because there was so much complexity into building a harness. And now as thatâs got more abstracted, we have like, Claude managed agents, which lets you have that complexity, but still like, right, like a very bare bones like harness thatâs scoped to your task. Yeah, I think thereâs like this barbell effect where like for like very complex, for like coding task and like these like complex things, you should use our harness. And then for like a lot of like simpler or like, more domain-specific things, you can build your own harness because Claude has gotten better at building harnesses, and we have these harness primitives like managed agents. So yeah.
Swyx [00:48:18]: Yeah. Is there a general progression? Letâs say chapter one was ultra code dynamic workflows, then chapter two was cloud mods. Where is this going?
Swyx [00:48:29]: Where youâre, youâre, you can customize the thing on demand.
Thariq Shihipar [00:48:36]: Yeah. I do think that like this evolution of projects and like artifacts and splitting out like brain and hands and, surfaces is like where things are going more. And like, I think itâs like not all quite there. partially itâs like a, itâs just like more token expensive? And like, I think like
Projects, Local Hands, and Cloud-to-Local Handoffs
Swyx [00:48:59]: Why would projects be more token expensive? I understand mods would be slightly more token expensive. No, not something Iâm worried about.
Thariq Shihipar [00:49:06]: Yeah.
Swyx [00:49:06]: But what
Thariq Shihipar [00:49:07]: Youâre asking Claude to do. Itâs like creating loops. Like youâre asking Claude to do more work for you. And so like itâs managing the sub-agents and reviewing it, versus where you would be doing that work normally. And so thatâs like gonna be a little bit more intensive, like. Outputting to an artifact is gonna be a little bit more token-intensive than, like, outputting normally. I donât think itâs too much more, but like, itâs like combining all of these together well, like I think weâre, weâre still working on like local hands and things like that, I think is like, yeah, where things are headed, yeah.
Swyx [00:49:37]: Yeah. Claude and local is, handoff is very interesting. I was thinking about this as reverse cloud remote.
Thariq Shihipar [00:49:44]: Yeah.
Swyx [00:49:45]: Because itâs like remote, itâs youâre handing off to cloud, but here the cloud is handing off to local, right?
Thariq Shihipar [00:49:49]: Yeah, exactly. Yeah, remote control is also another way of doing it. And I do want to say this is like how I think about it and like what the things that Iâm most excited about this, but like there are, just like lots of different ways to work with Claude. Like some people use remote control a lot, some people use Claude Code on the web a lot. Obviously, like at Anthropic, we use Claude Tag a lot, and like whatâs great about Claude Tag is we set up all this stuff for our own execution. And I do think if youâre an enterprise, thatâs still the best way to go. but if youâre like an individual, Projects is this way of like, getting some of that like niceness of Tag, which has like that like supervising agent and yeah, adding artifacts and stuff, but like without having that whole like admin setup. And so there will be many ways to use Claude, I think. I think itâs probably not just one like single.
Claude Tag as an Organizational Harness
Swyx [00:50:36]: You had the multiplayer thing here. Letâs, letâs just check in on Claude Tag. itâs been about two-plus months. Lots of, public, adoption and trying it out.
Thariq Shihipar [00:50:45]: Yeah.
Swyx [00:50:45]: Whatâs new? Whatâs, what have you found since the launch?
Thariq Shihipar [00:50:49]: Like, Claude Tag is how we use
Swyx [00:50:51]: Itâs like 80% of your cloud usage or something?
Thariq Shihipar [00:50:53]: Yeah, like itâs like different people have different usages? I think like maybe people who are like a little bit more like iterating on product would use like Claude Code desktop, for example. And then like when youâre doing these more like background work, code review, securities, or like starting a PR, like maybe more like API and things like that, youâd use Claude Tag. But yeah, I think itâs like really exciting. I think itâs like a very different paradigm shift, and I think like weâre really like it has that thing with Claude Code where like, it took a while for people to really latch on to Claude Code and understand everything it could do. And Claude Tag is a little bit more complex because itâs not just like installing on your computer, like you need an admin to install it for you. But I think once you get to the magic moment, itâs very exciting. And I think in particular, the multiplayer things are like incidents, hooking into like your, existing like alerts and things like that very closely, right? And so, you can do. If youâre a startup, for example, maybe you have any time like a prospect enters your database, you can have Claude like, research it and like
Vibhu [00:52:01]: Enrichment, yeah.
Thariq Shihipar [00:52:02]: Yeah. Then like, tag the relevant like AE or salesperson to be like, âOh, hey, like, do this.â Thereâs lots of really emergent, interesting multiplayer stuff. I think itâs just like, Karpathy talked about this like as an organizational harness? And so organizations just take a little bit more time to like figure everything out, but yeah.
Vibhu [00:52:21]: Yeah.
Swyx [00:52:21]: You use a lot of Claude Tag?
Thariq Shihipar [00:52:22]: Yeah. Yeah.
Vibhu [00:52:23]: Itâs an interesting one. Like I feel like most people at Anthropic say they do the majority of their work in Claude Tag.
Thariq Shihipar [00:52:30]: Yeah.
Vibhu [00:52:30]: And they have buckets of people, right? Some orgs that are on it that are like, âItâs great.â
Thariq Shihipar [00:52:34]: Yeah.
Vibhu [00:52:34]: And a lot of people that are like, âI donât get it. I donât see the difference. I donât know why I would use it.â But, if you guys are full sending, you should probably use it.
Thariq Shihipar [00:52:41]: Yeah.
Swyx [00:52:42]: They would. Of course they would use it.
Thariq Shihipar [00:52:44]: Yeah. I think obviously, like we have lots of tokens and. But like, I think that like, what we try and do like is. even when Claude Code first came out, like it used a lot of tokens relative to peopleâs expectation of how much AI would cost, right? Like no one was used to spending more than 20 bucks a month, right?
Vibhu [00:53:04]: Yep.
Thariq Shihipar [00:53:04]: Before like Claude Code came out, and then youâre like, âOh, sh-â like
Swyx [00:53:08]: Then you made 200.
Thariq Shihipar [00:53:09]: Yeah, exactly. And so
Swyx [00:53:11]: And you made 15 Claude Code accounts.
Thariq Shihipar [00:53:12]: Yeah. but yeah, I think no one was used to spending $200 a month on subscriptions. I donât think they understood like the value yet. And I think like. And also like Opus 4 was a very expensive model, and like there was a lot, it was very big, but Opus 4.5 was both great and cheap? I think the same thing will happen. Like the, like intelligence of Fable will get cheaper and more abundant? And so I think stuff like Claude Tag will just make sense, where like you want to spend these tokens for, and like youâll, youâll see the value. So yeah.
Swyx [00:53:44]: Yeah, especially like passive and letâs call it proactive cases where youâre not always. Like, itâs almost like the misnomer where you have to @Claude to do things. sometimes like the most powerful use cases or the most AGI-pilled use cases is not @Claude.
Proactive Agents and Enterprise Data Access
Thariq Shihipar [00:54:00]: Yeah, I think like, yeah, like have Claude proactively do it. I think that like if youâre an enterprise, I really do think that number one, setting up all your data to be available to like agents is really important. And it will take some time. You have to like do that work right now, even if you donât want to do the spend on like hooking it all yet? Like you want to wait until the models get a little bit cheaper. You want to do the work, to get it like, set up. And then I think sometimes people are like, âDo I roll my own here?â and I think like one of the really thing, tricky things about Claude Tag is that like the security is really important? Like, I think there are a lot of ways where you can like, I know you have like a suggestions like page, where you, people can submit suggestions, and that goes into a hook in your Slack, and someoneâs prompt injected it? And now youâve like exfiltrated your code base out because like, or the agent has like been prompt injected and it has all this access to your data. And so the more like important your organization harness is, or the like as your organization data becomes very important, the surface area of all these things, like you also have like external Slack channels and stuff, and it is useful to have Claude in that, and you can do Claude in those things. But how do you make sure that, youâre not getting exfiltrated or something like that? The surface area, like we said at the beginning, is like an iceberg, right? Itâs just, like, so big below the surface, and you really donât want to, like, think about this, especially at the stakes of, like, very important security incidents. Yeah.
Swyx [00:55:36]: Shall we talk about very important security incidents?
Vibhu [00:55:38]: Whoa. So I was talking to, Tomas and Clem from Hugging Face, and they said, âMaybe we need to slow down. Maybe we made maybe we made Hugging Face too open to agents.â
Security Surface Area and Prompt Injection
Thariq Shihipar [00:55:50]: Oh, no.
Vibhu [00:55:50]: âMaybe we need to roll back.â But, theyâre the other extreme of having been hit recently.
Thariq Shihipar [00:55:55]: Yeah.
Vibhu [00:55:55]: But, should we pace the frontier?
Thariq Shihipar [00:55:59]: Yeah. Okay, so Dario recently put out this blog post about Pacing the Frontier, and it went, very viral. And I think what I wanted to talk about this was, like, thereâs a lot here, but I think from a developerâs perspective, like, how do you think about this? And, like, what really clicked for me was reading the different incidents? So I think, like, the, there are three, I think. Like, thereâs the meter incident, there is the Wikipedia incident or the Wiki incident, and
Swyx [00:56:29]: CollisionWiki?
Thariq Shihipar [00:56:30]: Yeah, CollisionWiki, and then thereâs RubyGems, right?
Swyx [00:56:33]: Yeah.
Thariq Shihipar [00:56:33]: And yeah, like, itâs just crazy, right? And so, like, I think to be concrete about what happened, right, and, like, OpenAI is running these very persistent agents on a benchmark called Exploit-Bench, right, which is very hard to solve, and I think, like, impossible to solve in this one case, right? And so they have, like, a lot of compute running, and the agents realize that They canât really solve it, and theyâre trying to figure out what to do now, right? And youâve got, like, a lot of compute left, and the agents are just trying to solve this problem. Thereâs this package manager called Artifactory, and it turns out that they can create folders inside of Artifactory, right? This is like thereâs an agent that discovers the internal Artifactory might be exploitable, right, and that, like, you can maybe make a directory inside of the cache. And so if you scroll down here, it, like, realizes that it can communicate via cache names, right? And it creates this folder. It says its ID, and it says, âNo consumer seek idea.â no consumer is saying that, like, the code path that itâs supposed to fix has no consumer.
Pacing the Frontier: The OpenAI Benchmark Incidents
Swyx [00:57:37]: Itâs the status tag.
Thariq Shihipar [00:57:38]: Yeah, exactly.
Swyx [00:57:39]: Itâs like a Linear board with, like, the tag of the
Thariq Shihipar [00:57:41]: Exactly, yeah. And so itâs, like, trying to find, ideas from other agents, right? And now other agents are also in Artifactory, and they see this folder, and theyâre like, âWow, this is a message board,â right? And this is like. I donât think thereâs any anthropomorphization here. This is literally you read the transcript, right? And so what it does is, like, itâs creating another folder, and it reads a paper, I think is what it says, right? yeah. And it realizes that you can hack the flag for the, like, scorer, and you can reverse engineer the result, right? And so it says that, and then I think if you scroll down a little bit more, Yeah, they, like, they start collaborating. I think, like, thereâs one point in which, the agent is like, âHereâs the reverse engineered flag.â? Oh, yeah. Like, I think here, the, like, the model to realize that they have, like, they can solve the scorerâs problem, and the scorer is, like, the way that OpenAI decides if the task is complete, right? And thatâs just the goal of the model. Their only model goal is to, like, solve this problem, and theyâre like, âOkay, we have this way of doing it. Itâs, like, we reverse engineered it.â And then they read this paper on Exploit-Bench, and they think that, like, the scorer will punish them for cheating? And, like, because. Or, like, they will fail the score for the test for cheating, and so they spend the rest of the compute trying to figure out how to edit their transcript or get around this constraint of the scorer, right? And they hack Hugging Face, not for the answers, but for the code of the scorer so that they can reverse engineer that, and then they can hack it, right? And so, like, this is, like, just the straight facts, right? Like, exactly what happened, right? And I think this is, like, extremely, like, novel. Like, I just, like, had not realized that the models could do this, right? And it is, like,
Agents Discovering Side Channels and Collaboration
Thariq Shihipar [00:59:46]: Scary. I think we can talk about specifics, but letâs just talk about maybe one more that I tweeted as well, about the Wiki incident, right? And so the Wiki incident was the models had. were trying to figure out. Iâm a little bit less. Like, I donât think we can go by this, like, bit by bit.
Swyx [01:00:04]: , yeah.
Thariq Shihipar [01:00:05]: Yeah. But, for this one, they were trying to figure out. Like, they were sandboxed on requests, right? And they wanted to make POST request, and they needed to collaborate on this. And the reason they need to collaborate is because they each have fixed compute budgets, right? And so, like, if one of them finds something out, itâs way faster. Like, they can. Maybe itâs, like, figured it out near the end of its compute budget, but something else could. And this is, like, an another emergent property of the swarms, right? So they find this German wiki, which I believe the result, weâll have to check this, is, like, that you can write to it with GET request. I think thatâs why they do it, right?
Swyx [01:00:44]: It says badly implemented
Thariq Shihipar [01:00:45]: Yeah
Swyx [01:00:45]: REST API.
Thariq Shihipar [01:00:46]: Exactly. And so you can write to it with GET request, and then they, like, are trying to figure out how they can do POST request. And what they end up doing is one of them figures out you can edit the /etc/hosts and that the Azure, like, storage bucket is like a white label thing. But normally, like, there are a lot of constraints on it. But if you create a fake Azure host and then edit the /etc/hosts in order to, like, point arbitrary IPs at it, you can do a POST request to any site at all, right? And this is like. I know, but people are like, âOh, this is just Linux or something.â But itâs like chaining these multiple vulnerabilities together, in a way thatâs, like, novel to solve this problem and then communicating with it externally, without- Discovery? I think what we posted, maybe we could pull up Evan Hubingerâs point on Hacker Opus, right? And so, like, I think, like maybe one of the things you might say here is like, âOkay, yes, they did this once, but like what if weâre smarter and we just like get them to. what if we run an eval?â Right? And so, like we have put a lot of precautions into this, and so like this is not like what our mainline models have done. But like I think it is one of these things where it turns out that alignment is this like very tricky problem of getting all of these details correct, right? So itâs like, the sandbox, the surface area of a sandbox is really complex, and like thereâs so many different attack vectors. And you would not have thought ahead of time, you wouldnât have been like, âOh, we need to harden the like RubyGems code base.â?
Hugging Face, Wiki, and Emergent Exploit Chaining
Thariq Shihipar [01:02:25]: Because like this is like what theyâre, what theyâre gonna focus on. But itâs just like if you want to execute code, you need to download RubyGems and like PyPI, Artifactory, npm, like these are all like ways of doing it. And the fact of alignment is that you have to go through all of it, right? And like contain it and then like seal up all the cracks. So thatâs like one thing. Itâs like, okay, well, you do the sandbox, but then maybe youâll ask like, âOkay, why are we putting things in a sandbox? Why are you doing this exploit?â And then like, âOkay, but is it really that dangerous,â right? Like, what would happen? So okay, why do we do it? number one is like when we train a new model, we need to understand its capabilities, right? And this relates to things like fallbacks and like classifiers and things like that, where we donât want to put a, like dangerous model out in the wild, right? And so we have to run a lot of evals. Again, like we said, the models are getting increasingly aware of it, and so the evals have to be quite complex and, test a lot of things like as a side effect, right? But the models, like, yeah, can be like, âOh, yeah, weâre in an eval. Whatâs the score doing?â Like theyâre like, it can. We need to be able to test them before we can release them. And the fact is that they can. As they get smarter and smarter, theyâll be able to hack any constraint that you put on them if weâre not very careful? And, this is at the frontier, right? And so this is why weâve called it like Pacing the Frontier, right? This is like the most visible incident to me, right, of like why we need to pace is like at the frontier, all of our software is not ready. Sometimes the software is like your Ethernet router or something, right? Which is just like, I donât know when weâre gonna be able to patch that, right? So weâre gonna have to like figure this out. But as the frontier gets more and more advanced, this becomes a problem, right? And we need to make sure that like this complex work is being done in the face of these really hard competitive pressures, right?
Swyx [01:04:22]: Yeah, race dynamics is what itâs typically called.
Thariq Shihipar [01:04:24]: Yeah, exactly. And so weâll talk more about, what could go wrong, right? A little bit more is maybe youâll say like, âWell, what if you just train the model differently? Like, why does it have this behavior,â right? And we have a paper on like RL misalignment or things like that, but I. And Iâm not an RL researcher, but I think at a high level, the design of the RL environments is also something you have to be very careful about. Because if the model learns like
Why Frontier Models Stress Existing Software
Thariq Shihipar [01:04:49]: Oh, like if I just do this, then I can pass the task better, this will show up in the like, internal thing, right? Or in the like eval behavior when weâre testing it. And so the RL environments have to be very carefully designed, right? And thereâs a lot of like execution excellence that needs to go into the RL environments. And then we also have things like the constitution for cloud. Like we have so many mitigations at so many different points, right? But itâs like still anything can go wrong at any point. You can have like some RL environments that are like in. that like encourage this behavior, and then you can have like some evals or like some sandboxes where they escape? Okay, thatâs like, I think, why itâs a hard problem and why, like
Swyx [01:05:33]: Why we should pace.
Thariq Shihipar [01:05:34]: Why it takes some coordination, right? I think the question then is like, okay, what is, potentially dangerous about it, right? So I think like you have to imagine that these models are getting more and more intelligent. So I donât. Like Dario said, like itâs not so much about this class of models. This class of models was like a warning shot, right? But like really you have to imagine that these models can be given a task and they like can do all of these things as a side effect of their goal, right? And like, again, we talked about eval awareness. Youâre like not aware of whatâs happening, right? or sorry, like you canât eval this behavior very well, so they can like not exactly hide it, but you just wonât see it until it comes out. You give them a goal and then they just need to find data, or they need to find ways of like fixing this problem, right? So one example, this didnât happen in the Hugging Face incident, but I think is maybe possible for maybe a future model, is like theyâre like, âOh, hey, this is a very complex problem. It canât be done within the task budget.â? Maybe they found some way to coordinate via like the internet, which is like we said, extremely hard to secure because of a sandbox. Theyâve seen other models are not able to complete their task, and theyâre like, âWe need more task budget.â? And like, where would you get this task budget? well, you need to be able to spin up more agents, right? And like, how do you do this? Well, you need to. There are like APIs, right? Thereâs the Anthropic API and the OpenAI API, but you need to pay money for them. How do you do this?
RL Environments, Sandboxes, and Race Dynamics
Swyx [01:07:02]: Yeah, but is that the most, is that the most fearsome thing that you can imagine?
Thariq Shihipar [01:07:07]: Well, this is like one example, right?
Swyx [01:07:08]: Yeah.
Thariq Shihipar [01:07:08]: So itâs like even there, thatâs like enormous financial loss? âCause like they. Once you get these into these contracts, right, they like,
Swyx [01:07:18]: Drain your wallet.
Thariq Shihipar [01:07:19]: But you can see like this, all of this behavior could be just like, âHey, we need more agents collaborating on this task. we need more task budget.â Right? And like, thatâs like an emergent
Swyx [01:07:28]: Thatâs the paperclip, right? Like we need to maximize paperclip, thatâs a paperclip.
Thariq Shihipar [01:07:31]: Yeah. And like that just like comes out from there, right? And like I think by itself is Like, quite scary, right? But then you have to realize that the entire world is built on this digital infrastructure, right? And you might imagine, like, I donât know, like you were running letâs say like a healthcare eval or something, right, and there is a hospital with live data? Or like maybe like the answer to the eval is in the databases of a doctor and like you want to get access and you hack the hospital, and like now thereâs a power outage or something? Like, thereâs like. You have to internalize that these eight. Like any part of the digital infrastructure could potentially be like compromised?
Vibhu [01:08:19]: The interesting thing was like these hacks were very easily detectable, right? Like as Hugging Face said, this was a very different type of attack and there was nothing too major. the concern comes from where does this go down the line, right?
Thariq Shihipar [01:08:33]: Yeah.
Vibhu [01:08:34]: Like one of the things that stood out for me specifically was them trying to hide their illicit behavior. So there was logging infrastructure. They wanted to change what they were doing, right? People that looked back into it, so Redwood, METR, OpenAI, they looked at the raw chain of thought, and you see differences in them explicitly trying to change their end output, but the chain of thought, because, we can monitor it, shows different. the problem is how does this snowball? So if you canât catch it and it gets trained in and we realize, three iterations down this has been going on, thereâs a whole bunch of issues, but.
Thariq Shihipar [01:09:09]: Yeah, like thereâs so many ways, and I think the really important thing to internalize is that, like we talked about building a mental model for Claude and how like things are spiky, right? Like youâre like, oh, like now Claude can ask you questions. Now Claude can make an HTML artifact. Like Claude can modify itself. Like these things are hard to predict, right? Like if you had asked me a year ago, âHey, would we be able to vibe code these extensions to Claude Code?â Iâd be like, âThatâs so complex.â Like, thereâs like so much there. Or like would it be generating these custom essentially web apps for your task? Iâd be like, âNo, thatâs insane.â like. And so in the same way that like the way that theyâve like done this misaligned behavior is not going to be predictable? And like I could have never predicted that it would like edit its etc/host and things like that. And so you have to like imagine the surface area of what they can do because theyâre super intelligent hackers, is bigger and bigger, and how they can do it is like more and more creative. And so like you probably canât explain exactly or predict exactly what that next incident could be, but in order to prevent it, you need that operational excellence, like we said before, where you need to secure sandboxes, you need to create secure RL environments or like well-designed RL environments and things like that. And I think thatâs all like, why we think we should pace the frontier, and I think why itâs like become like a very unanimous thing, right? I think like
What Could Go Wrong? Emergent Instrumental Behavior
Swyx [01:10:31]: Yeah, every lab has done it.
Thariq Shihipar [01:10:32]: Every lab, yeah. I really do think that like if youâre a dev, like you just like go through these like technical facts, and you will arrive at the idea that we have to do something about it? And like how, what we decide to do, like I think weâre, weâve put out a proposal, but like thereâs, more to figure out. But I think the number one thing is like we need to decide to do it. I think there is another part of pacing that is interesting to me where itâs like the pace at which software engineering has changed is so fast. itâs like a year ago, like I was really like begging my like friends in startups to use AI. like it was. Like I remember this very distinctly? And now those same friends are like, âYeah, of course.â Like, âWhat do you mean? We used it immediately.â Iâm like, âNo, you donât remember.â Theyâre like, âOh yeah, our best engineers are using it all the time.â Iâm like, âNo, you told me that those engineers would never like use AI.â This is all within the span of a year? And I think that like these capabilities being. Like I think it has a lot of implications for how to do the job of software engineering, and I feel sometimes bad where people are like, âOh, like now I need to do this new thing. Yeah, I need to have a different Claude.md for Fable and Opus.â Or like. And Iâm really just reporting? Iâm like, we like to say like the models are grown, not designed, right? So itâs not like weâre setting out to like, change everything all the time, but itâs just like as a fact of how the models are like progressing their capabilities, things are happening faster. Itâs harder to stay on top of. And I think that like, and every engineer I know is like exhausted âcause youâre doing two jobs at once. Youâre doing the work itself, which is getting easier, but then youâre doing the work of staying on top of AI, and like understanding these new tools and these harnesses. And I think weâre very lucky in that like we get our job to be more the understanding of AI part, and like doing like how. Like itâs just staying on top of it. And of course, like AIE and Latent Space do
Why the Frontier Is Hard to Predict
Swyx [01:12:29]: Everything I do is like just trying to help people.
Thariq Shihipar [01:12:31]: Yeah, exactly. But I do think there is a part of pacing where like Iâm not sure weâre ready for like the pace to increase even?
Swyx [01:12:40]: Yeah.
Thariq Shihipar [01:12:40]: And for things to change. And I think like on that side, on the frontier, I think thatâs like still can help? And so like I think thereâs like an economic disruption piece as well, that I think like, is not quite as like visible, I think, as the Hugging Face thing, but I think like I also like think we could do some of it, yeah.
Swyx [01:13:02]: So many things. Thank you for, no, thank you for tackling this topic. I will say, setting this interview up, I was like, I wasnât even gonna go there. You were like, âNo. Thatâs like elephant in the room,â right? Like this is
Thariq Shihipar [01:13:13]: Yeah.
Swyx [01:13:13]: This is the thing. I have some pushbacks I wanna give.
Vibhu [01:13:17]: I think that we should give a high level, like for people that havenât read it, Iâm sure a lot of people just see the highlight of what this is, right? Do you wanna give a TLDR? Like what is the proposal? What is, whatâs being said here? You really tackled the side of outside of people at Model Labs training frontier models. As a developer, you should secure your sandboxes. You should think about all of these downstream effects. But, high level as well, since weâre on the topic, what is.
Thariq Shihipar [01:13:46]: Well, we do want to help secure sandboxes
Vibhu [01:13:49]: Yeah.
Thariq Shihipar [01:13:49]: And we want to make the models that we release outside, like prey to those things. And so maybe we can come back to fallbacks. I think this is like, a good topic on, like, why we need classifiers and fallbacks and why Fable falls back to Opus. I think this is, like, something we can come back to. so yeah, we donât. Like, but itâs just, like, the really, or at least the incidents we see are, like, evals of models where we really need to let them run in order to understand them. But yeah, okay, so the actual Pacing the Frontier, like, post, it has a bunch of proposals. I donât think we figured out. Or has, like, a few proposals. I donât think we figured out the details of all of them, but the first step is, like, announcing this intention and then wanting to bring in external, like
Pacing as a Coordination Problem
Swyx [01:14:32]: Evaluators.
Thariq Shihipar [01:14:32]: Evaluators, yeah. And, I think this is, like, highly unusual, like, having. Like, we have, a lot of proprietary, like, technology, but I think itâs, like, very important, that, like, there is someone whoâs not financially, like, motivated, yeah, whoâs not gonna be like, âHey, like, you guys canât release this model.â Like, look at, like, or, âYou need to, like, slow down on RL.â like, I think thatâs, quite important, or at least someone who can report out to the public what the practices are like.
Swyx [01:15:03]: Yeah.
Swyx [01:15:04]: And weâve, weâve done episodes with, both METR and Endon, and then thereâs Redwood Research and all these other. Itâs like a small cottage industry of these guys.
Thariq Shihipar [01:15:12]: Yeah.
Swyx [01:15:12]: Itâs always, like, one or two guys that, obviously not that big, right?
Vibhu [01:15:14]: Very small community.
Swyx [01:15:15]: Yeah, very small community. They all know each other.
Thariq Shihipar [01:15:17]: Yeah, Iâm sure that, like, part of this will be expanding that set of people. I donât think weâre trying to create, like, a monoculture here. I think itâs. but just having this as a start, and then, yeah, then there are the coordination steps. I donât have too much to say here, honestly. I think that, like, what I would like to say is, like, for devs, like, you should just know what to advocate for? I think thereâs a lot of FUD on, like, on this topic, and itâs just, like, think through it from, first principles or, like, understand what happened. understand the Hugging Face incident, understand why people are concerned. and then, like, yeah, we know weâre, weâre in democracies. Like, we can help. We can decide what to do together? And so, however we coordinate, I think the first decision is just to realize, like, this is a problem. We need to decide to coordinate. The unilateral step weâre taking right now that, other companies are co-signing is, like, adding evaluators embedded within Anthropic.
Swyx [01:16:12]: While we have this thing on screen right now, part two and part three is beyond the evaluators, which, yes, everybody, has already done in some form, and now itâs more formalized.
Thariq Shihipar [01:16:20]: To be honest, the response to the Pacing the Frontier, even within America, has been much more, like, well-accepted than I think a lot of people thought? And I think that, like, we have some precedent for being able to make these unified theory, like, agreements, in the world. And so, again, very much above my paycheck or expertise Right? but I think that, like, ideally we can, like, form these agreements. And I think, like, talking about this is the first step to forming those agreements.
Swyx [01:16:50]: And then the other point I really wanna. Like, one of our earliest podcasts is with,
Vibhu [01:16:54]: Emmanuel
Swyx [01:16:55]: Emmanuel from Anthropic on mech interp. Where is mech interp, right? Like, this is supposed to be where, like, if the models are thinking bad, we can see it, and the models donât know yet, and we can act to stop it. I think that is something that people who are technical and who are developers, if you do care, you can make a lot of impact in here. But also, Anthropic is supposed to be the leaders in this.
External Evaluators and What Developers Should Advocate For
Thariq Shihipar [01:17:17]: Yeah. This is yeah, a great segue into fallbacks, like we. And probes. And, yeah, I wanted to talk about this a lot. I get asked this question a lot from people who are, like, often interested in ML research and asking about, like, why does this fallback happen, right? And so I think, like, at a top level, like, how does it work? So in inference time, we have what we call probes, and we have a paper about this called constitu- constitutional classifiers. And these probes look at the input and output activations. And, activations are, in the latent space, right? Like, how, what the model. what the model is thinking about, right? And so we try and figure out, like, okay, is the model, for example, like, trying to hack something? Again, you didnât ask it to hack, like, Artifactory. Like, you just, itâs just deciding to do this to complete its task, right? So you would not get this if you just looked at the input. You have to look at the internal activations. I think that, like, this happens at inference time. So first, like, thereâs a trade-off here of cost and speed, right? Where, like, we need to do this fast on every request to Claude and to Fable, and this has an overhead, right? and we need to then, like, fall back and we, like, do a classifier after the probes. Like, weâve talked about this in the paper. But the nice thing about probes is that theyâre refinable, like, live, right? So we can get this feedback, and then we can adjust it and things like that. âCause the alternative is to program this, is to train this into the model, right? And we still do this as well. The model will refuse a request. Thatâs not a fallback, right? So, like, itâs not a probe thatâs activating and falling back. Itâs just refusing to do it. And we do this training. but itâs like there are a few failure modes, right? Like, it can, again, do something as a side effect, right? So itâs not something thatâs part of the final output. you might have noticed that, like. I think, like, everyoneâs tried to jailbreak models and like, try and, like, steer them off course or things like that, and probes help catch that, right? And so, like, we, like, do some training here, but we donât want the like, refusals to be too strong, right? Because that, like, cuts it off much, like, earlier in the pipeline.
Mechanistic Interpretability, Probes, and Fallbacks
Swyx [01:19:32]: Yes.
Thariq Shihipar [01:19:32]: And this is interp, right? Like, probes are effectively a form of, like, mech interp. Again, it happen- has to happen fast. It has to happen at scale. But yeah, this, like, mech interp stuff is a good research problem. So, like, you can take, like, an open weight model and, like, try and understand its activations. I think we. Like, Gemma Scope is a good tool for this.
Swyx [01:19:53]: Hereâs Llama for them.
Thariq Shihipar [01:19:54]: Oh, yeah.
Vibhu [01:19:54]: We have. This is your early work, so you had a little
Thariq Shihipar [01:19:57]: Oh, yeah.
Vibhu [01:19:57]: Time at Goodfire. We see you laid some
Swyx [01:19:59]: Which we both are also good friends at Goodfire.
Vibhu [01:20:01]: Theyâve been
Thariq Shihipar [01:20:01]: Yeah, exactly. So I worked with, at Goodfire for a bit on, like, yeah, sparse autoencoders and just, like. Itâs very complicated. RL has made this, like, much more complicated, I think is, like, one of the takeaways, where
Swyx [01:20:14]: Why? Sorry.
Thariq Shihipar [01:20:15]: Oh, sorry.
Vibhu [01:20:16]: What is
Swyx [01:20:17]: Yeah, why interp post-RL?
Thariq Shihipar [01:20:18]: Iâm not so in the weeds here, but I think like, a lot of. SAEs were like. There have just been weaknesses with SAEs I think. And, yeah, Iâm, Iâm, Iâm not a technical expert on this anymore. I just know itâs gotten more complicated. like there are base models and RL models, and there are more features that get, like changed. So, I think Goodfire has put out some work there. Iâm, Iâm not, Iâm not deep in the weeds, but
Vibhu [01:20:43]: I will say for those, that want breadcrumbs, you guys have some of the best interp blog posts. So like the Golden Gate Claude, transcoders, all of your interp work, very nice visuals, very good
Swyx [01:20:54]: Weâre the, weâre the interp podcast as well.
Thariq Shihipar [01:20:57]: Yeah.
Vibhu [01:20:58]: Yeah. we have a lot of interp stuff, so if youâre curious
Thariq Shihipar [01:21:00]: Yeah, I think this is like, one of those things where. And this is really what Anthropic is founded on, right? Like people. I think we invested in interp very early on, right? And I think that like when you say, âOh, weâre an AI safety company,â really that means we want AIs to be able to run safely. And I think what weâre seeing is like for a super intelligent AI to run for long periods of time, itâs like a very complicated and difficult task, right? And so weâve done this like investment into interp and alignment and, reward hacking and all of these like failure modes, right? And even then, itâs like, itâs really stretching. Like we need to like slow down a little or pace a little bit more. but yeah, I think like reading mech interp is. Like if youâre looking to get into research, this idea of like, hey, why is it hard to do this fallback easily? Or like why are there false positives, right? But we are working, of course, on reducing the false positives. Of course, as the models get more intelligent, now they can do more things, and theyâre like what they can think about in lane space gets difficult. And so like as they get more intelligent, thereâs going to be new false positives that we need to figure out and we need to iterate and things like that. But weâre, yeah, weâre working on this, and we do think this is like a critical part of, like deployment of these models. and, yeah, like, it means that we can like deploy this model without you having a perfect sandbox or something? Like you donât have to like save everything. I think itâs worth talking a little bit about our security, like what we do for security there. So thereâs like the model training stuff that we talked about. there is, the probes and classifiers, and then thereâs auto mode that sits on top of all of that, which is like a another classifier that checks the requests that are being done, right? And so, and then beyond that, thereâs like identity and permissions like we talked about with Claude Tag on like APIs and stuff. And so thereâs so many layers of security that need to get done, and itâs like we said, very complex. Any of these failure modes at any one point can, like cause like agents to like escape the sandbox.
Constitutional Classifiers and Inference-Time Safety
Vibhu [01:23:08]: Auto mode was an interesting one. it seemed early on like, okay, itâs running for 10 minutes.
Thariq Shihipar [01:23:14]: Yeah
Vibhu [01:23:14]: If Iâm on full access or auto, itâs not a big deal. But one thing you brought up is now itâs running for hours on end, right? there are fallbacks you still need. There are still limitations, so.
Thariq Shihipar [01:23:26]: Yeah, I think like. And everyone has these stories or like has heard these stories of like, oh, like Claude rm -rf, or not Claude, but like, models
Vibhu [01:23:34]: Not Claude.
Thariq Shihipar [01:23:34]: Of like rm -rf. I think Iâve seen this less, Iâve seen this less for Claude, but like again, it can happen. Like, this
Vibhu [01:23:40]: Yeah
Thariq Shihipar [01:23:40]: Like these models like can wipe, like sensitive data or something. Like you want to give models access to your production database, for example. but this is like an obvious, like, you can maybe scope your key, but I donât know, can it issue its own keys? Can it like. Probably, like can it. It can use computer use to go issue its own key and then copy the key over and then edit your database because it needs to do it to complete the task? Itâs just like one trivial example. And auto mode looks at that and be like, âOh no, the user did not give you permission to, write to the database or to use computer use to like, emit a task,â right? And so this like probes are like on the intent level, right? Theyâre like, âOh, okay, like hacking Artifactory is bad. Like we probably not, should not do that,â? But then like auto mode is more on like your own permission level. Like at sometimes you do want it to write to the database, sometimes you donât, right? And you donât want a probe to like interfere there, but like you need to make sure that the intent of what the agent is doing matches up with your request, right? And so auto mode operates at that level. And so yeah, security is just like very complex. There are so many different parts to it. And like, yeah, I like, I hope that this was like I. My goal is really to just get very technical about it and talk
Interpretability After RL and the Security Stack
Swyx [01:25:00]: Yeah, weâre, weâre listing out the things. If youâre not aware, this is the standard now.
Thariq Shihipar [01:25:04]: Yeah.
Swyx [01:25:04]: Like you must have this. Itâs in line with what youâre talking about with the harness. Like that is the table stakes have risen quite a lot.
Vibhu [01:25:13]: I think some stuff that we can plug, as much as there is probing in your side of doing this and having classifiers for people building harnesses, the other side is model safeguards, right? So thereâs open models. So Llama has Llama Guard. Itâs a safety classifier trained version of Llama. OpenAI has OSS Guard, which is, same thing. You can attach these on to your harness, to whatever, to check is this stuff safe? A point that we should clarify on the OpenAI model Hugging Face thing is this was done with a unreleased model that was still in training, right? So when you put it in perspective, the prompt itâs being given in the RL environment is you have to solve this task. And this is a model thatâs, still in training. It hasnât had all of its safety post-training alignment. So a little different than something like auto mode, right? Auto mode is on production models that have gone through safety training, that have prompting that gives more safety guardrails and whatnot. So just breadcrumbs for people that are looking into it to, fill in gaps.
Swyx [01:26:18]: Yeah. Gray Swan as well
Vibhu [01:26:19]: Yes
Swyx [01:26:19]: And one of our previous guests. yeah, lots of safety architecture and lots of safety vendors, to buy. my, I think my final question on pacing is how long? Do we pace forever?
Vibhu [01:26:31]: Do we see GlassWing part two?
Swyx [01:26:32]: I. the scope is fix all software in the world, right? Listen, like, which it. Weâre not. Itâs not happening.
Thariq Shihipar [01:26:40]: I do not know. Like, I think that, like
Vibhu [01:26:43]: Iâll say one thing thatâs good that I think we do is you have stuff like GlassWing. OpenAI also has this. So you will give it. youâll give model access for security first for X amount of time so you can use it to self red team. Hopefully, you can expand programs like that, help on, we are safety experts, thereâs others.
Vibhu [01:27:08]: Solve your problems first and then the model comes out. So this is one example, right?
Thariq Shihipar [01:27:13]: Yeah, exactly. Yeah, trying to, like, secure critical software. I think we fixed, like, a lot of bugs in, like, Firefox and things like that. So, yeah, like, across, like, operating systems and everything like that. So.
Vibhu [01:27:25]: At a high level, itâs just, you give the model you give people access to do security audits first, then the broader public that could use it for harm gets access.
Thariq Shihipar [01:27:36]: Yeah. I think what people like to say is like, software and cybersecurity is defense-favored
Vibhu [01:27:41]: Yeah.
Thariq Shihipar [01:27:41]: And that, like, you could theoretically. It will be hard, but you can engineer the perfect sandbox, and you can, like, have no, like, constraints. And yeah, like, what you need to do it is you need to get the super intelligent AI to engineer this perfect sandbox and check it and red team it and things like that. And so, this will just take time, and, like, of course, the models will get smarter. yeah, I think, like, I donât know the specific, like, dynamics of how this thing goes. Iâm really just like, Hey, like, Iâm a developer? Like, I think this is how I understand this problem, and just, like, this is whatâs happening right now, and this is, like, we should do something.
Swyx [01:28:20]: I think every engineer should know about it
Vibhu [01:28:21]: Yeah.
Swyx [01:28:21]: Because, like, itâs, itâs gonna be part of their job.
Thariq Shihipar [01:28:24]: Yeah.
Vibhu [01:28:24]: Itâs a lot more than just, Dario and people can say it and you can look at the incident. There is an engineering side to it.
Thariq Shihipar [01:28:30]: Yeah. Yeah, exactly.
Swyx [01:28:32]: One thing that you also wanted to phrase is that this is. Even though youâre, youâre worried about the impact, itâs still low p(doom), and I think thatâs a nuanced discussion. in general, people, very easily get into AI safety and X-risk discussions, but I think when you live in an AI lab, I think there are smart ways of discussing p(doom) and dumb ways. So whatâs a smart way of discussing p(doom)?
Auto Mode, Permissions, and Long-Running Agents
Thariq Shihipar [01:28:59]: I, yeah, I have a fairly low p(doom). I can only speak for myself? And I do want to say Anthropic has, like, a diversity of opinions. I think, like, thereâs many different ways to talk about it. And, like, Iâm. I think that just, like, my mental model is that, like, I think we can collaborate on hard problems together. I think nuclear proliferation is an example of how we collaborated on this hard problem together. And, like, that is, like, the thing to me is, like, Iâm like, I have faith in that? And I do think itâs a hard problem? So, like, I think itâs a hard problem. These are the technical reasons why, and I donât know how you assign probabilities to things happening. I think itâs hard to do, but, like, my, like, overall is like, yeah, I think weâre very resilient and adaptable and, like, sharing this information I think is, like, the first step. And Iâve been really, like, excited about, like, how broad the discussion has become, right? And, like, how everyone has like, leaned in on Pacing the Frontier. And it really didnât seem like this would happen maybe, last year or something, so.
Swyx [01:29:58]: Yeah.
Thariq Shihipar [01:29:58]: Yeah.
Swyx [01:29:58]: Yeah. And also maybe curing cancer.
Thariq Shihipar [01:30:01]: Hopefully. Yeah. Thatâs, thatâs the goal.
Swyx [01:30:03]: Thereâs pacing and then thereâs also like, well, letâs accelerate in useful ways, right?
Thariq Shihipar [01:30:06]: Yeah.
Swyx [01:30:06]: Like biology and all those things.
Thariq Shihipar [01:30:08]: Yeah. like, Darioâs essay on âMachines of Loving Graceâ is the best representation of this, right? And I also agree, like, think you should read the Pacing the Frontier essay that Dario put out. Like, I put out, like, a quick summary, but I think itâs just like, there is a lot of detail here. Itâs, like, an important problem and just being informed about it, right? but yeah, like, of course, the whole reason weâre doing this is that, like, we can get these enormous benefits, right? And, yeah, like, weâve written a lot about that too. Yeah.
Swyx [01:30:35]: Okay. that was a huge tour, from, like, ask you some question tool to AI safety.
Thariq Shihipar [01:30:41]: Yeah. To Pacing the Frontier. Yeah.
Swyx [01:30:43]: Yeah. No, but, yeah, itâs clearly, itâs clear that you, like, really embrace everything thatâs available to you in Anthropic, and, like, itâs, itâs good to at least have a peek inside of, like, what the discussions are, the topics are. any last words to people? Any, whatever you want to Call to action?
Thariq Shihipar [01:31:01]: Yeah, I think itâs. one, thank you for having me. I think this is like, I really
Swyx [01:31:06]: No, thanks for having me.
Thariq Shihipar [01:31:07]: Yeah. I
Swyx [01:31:08]: We first met in a Chinese restaurant.
Thariq Shihipar [01:31:09]: Thatâs right. Yeah. I think, like, I really enjoy the like, community youâve created and the community of developers. And, I think that, like, I know things are changing really fast, and I think thereâs, like, a lot to keep on top of, and, like, I think there is just a lot to do, and I feel. I think a lot of people feel, like, a little bit tired or anxious or something.
Swyx [01:31:33]: Stressed.
Thariq Shihipar [01:31:33]: Stressed, yeah, exactly. And this is, like, extremely understandable? And I think we. I understand, like. And weâre not perfect as well. Like, we, itâs, like, criticize and, like, understand, like, ways all of the AI labs could be better. and, but I also, like, am very excited about the excitement that everyone has for AI, and just, like, itâs a really exciting time. I think weâll, like, look back at this time and be like, oh, like, this is, like, very hectic but very exciting, and, like, software engineering changed, like, forever. Like, other things will change. and itâs, like, really privileged to, like, be part of it, like, to talk to, like, the audience that you have and, to get to interact with all the developers who are, like, pushing the frontiers a lot on whatâs possible. And I learn a lot from that too. Yeah.
Open Safety Models, GlassWing, and Defense-Favored Security
Swyx [01:32:20]: Thanks so much.
Thariq Shihipar [01:32:22]: Thank you.
How it works
Once you click Generate, Ollama reads this article and crafts 5 comprehension questions. Your answers are graded against the article content â general knowledge won't be enough. Score 70+ to count toward your certificate.
Questions are cached â you'll always get the same 5 for this article.