How Claude Code Works - Jared Zoneraich, PromptLayer

So, welcome to the last workshop. Um, you made it. Congrats.

Thank you.

Out of like 800 people, you're the last standing. Uh, it's the very, very dedicated engineers. Uh, yeah, so this one's a weird one. I got in trouble with Anthropic— with Anthropic on this one. Uh, obviously because of the title.

I actually also gave him the title and I was like, "Do you want to change it?" And he was like, "No, I'll just roll with it." It's kind of funny. Uh, uh, so, so yeah, this is not officially endorsed by Anthropic, but we're hackers,right?

And Jared is, like, super dedicated. He's, um, and the other thing I also, like, really enjoy is featuring, like, notable New York AI people,right? Like, so don't take this as, like, you know, this is the only thing that Jared does.

He has a whole startup that you should definitely ask him about. Um, but, like, you know, I'm just really excited to feature more content from local people. So, yeah, Jared, take it away.

Thank you very much. Thank you very much. And what an amazing conference. Very sad we're ending it, but hopefully it'll be a good ending here. Um, and yeah, uh, my name is Jared. Uh, this will be a talk on how Claude Code works.

Again, not affiliated with Anthropic. Uh, they don't pay me. I would take money, but they don't. Um, but we're going to talk about a few other coding agents as well. And kind of the high-level goal that I'll go into is, me personally, I'm a big user of all the coding agents, as is everyone here.

And they kind of exploded recently. And as a developer, I was curious what changed. What made it finally— what made coding agents finally be good? So let's get started. I'll start about me. I'm Jared. You can find me, I'm JaredZ on X, on Twitter, whatever.

Um, I'm building the workbench for AI engineering. So, uh, my company is called PromptLayer. We're based in New York. You can kind of see our office here. It's like a little building, so it's blocked by a few of the other buildings.

So we're a small team. We launched the product three years ago. So, uh, long for AI, but small for everything else. And, uh, yeah, what kind of our core thesis is that we believe in rigorous prompt engineering, rigorous agent development, and we believe that the product team should be involved, the engineering team should be involved.

We believe if you're building AI lawyers, you should have lawyers involved, as well as engineers. Um, so that's kind of what we do. Uh, processing millions of LLM requests a day. And a lot of the insights in this talk come from just conversations we have with our customers on how to build coding agents and stuff like that.

And also, feel free throughout the talk, we can make this casual. So if there's anything I say, if you have a question, feel free to just throw it in. Uh, and I spend a lot of my time kind of dogfooding the product.

It's kind of weird, the job of a founder these days, because it's half, like, kicking off agents and then half just using my own product to build agents. And it feels weird, but it's kind of fun. And, uh, yeah, the last thing I'll add here is I'm a big enthusiast.

We literally rebuilt our engineering org around Claude Code. I think the hard part about building a platform is that you have to deal with all these edge cases and, oh, uh, we're uploading data sets here, it doesn't work.

And you could die a death by a thousand cuts. So we made a rule for our engineering organization, if you can complete something in less than an hour using Claude Code, just do it. Don't prioritize it. And we're a small team on purpose, but, uh, it's helped us a lot.

And I think it's really taken us to the next level. So I'm a big fan. And let's dive into how these things work. So this is where, as I was saying, the goal of this talk. First, why have these things exploded?

What is the— what was the innovation? What was the invention that made coding agents finally work? If you've been around this field for a little bit, you know that, uh, a lot of these autonomous coding agents sucked at the beginning.

And, uh, we all tried to use them, uh, but it's, it's night and day. Uh, we'll dive into the internals. And, and lastly, we'll, like, everything in this talk is oriented around how do you build your own agents and how do you use this to do AI engineering for yourself.

So let's just go, uh, talk about history for a second here. How did we get here? Uh, everybody knows we started with, uh, you remember the workflow of you just copy and paste your code back from ChatGPT back and forth.

And that was great. And that was kind of revolutionary when it happened. Uh, step two, when Cursor came out, if we all remember, it was not, not great software at the beginning. It was just the VS Code fork with the command K.

And we all loved it. But, uh, now, now we're not going to be doing command K anymore. Then we got the Cursor assistant, so that little agent back and forth. And then Claude Code. And honestly, in the last few days since I made this slide, maybe there's a new version we could talk about here.

And, uh, at the end, I'll talk about, like, kind of what's next. But this is how we got here. And this is really, I think the Claude Code is kind of this headless, uh, not even this, this new workflow of not even touching code.

And it has to be really good. So why is it so good? What, what was, uh, what was the big breakthrough here? Let's try to figure that out. And again, throw this in one more time. These are all my opinions, uh, and what I think is the breakthrough.

Maybe there's other things. But simple architecture, I think a lot of things were simplified with how the agent was designed. And then better models, better models, and better models. Uh, I think the a lot of the breakthrough is kind of boring in that it's just Anthropic releasing a better model that works better for these type of tooling calls and these type of things.

But the simple architecture relates to that. So we can dive into that. The architecture, and, and this is our little, you'll see, uh, prompt wrangler is our little mascot for our company. So we made a lot of graphics for these slides.

But, uh, basically, give it tools and then get out of the way is what, uh, one-liner of the architecture is today. I think if you've been building on top of LLMs for a little bit, this has not always been true.

Obviously, tool calls haven't always existed. And tool calls is kind of this new abstraction for JSON formatting. And if you remember, the GitHub libraries like JSON former and stuff like that in the olden days. But give it tools, get out of the way.

Uh, the models are built for these things and being trained to get better at tool calling and better at this. So the more you want to over-optimize, and every engineer, uh, including especially myself, loves to over-optimize. And when you first have an idea of how to build the agent, you're going to sit down and say, "Oh, and then I'm going to prevent this hallucination by doing this prompt and then this prompt and then this prompt."

Don't do that. Just a simple loop and get out of the way and just delete scaffolding. And less, less scaffolding, more model is kind of the tagline here. And, you know, this is, uh, the leaderboard from this week.

Obviously, these models are getting better and better. Uh, we could have a whole conversation, and I'm sure there's been many conversations about, is it slowing down? Is it plateauing? It doesn't really matter for this talk. We know it's getting better.

And they're getting better at tool calling, and they're getting better optimized for running autonomously. And don't this is, I, I think Anthropic calls this, like, the AGI pill to way to think about it is don't try to over-engineer around model flaws today, because a lot of the things will just get better and you'll be wasting your time.

So here's the philosophy, the way I see it, of Claude Code. Ignoring embeddings, ignoring classifiers, ignoring pattern matching. We had this whole RAG thing. Actually, Cursor's bringing back a little bit of RAG in how they're doing it, and they're mixing and matching.

But I think the genius was Claude Code is that they, they scratched all this and they said, "We don't need all these fancy, uh, paradigms to get around how the model's bad. Let's just make a better model and then let it, let it cook."

And, uh, just leaning on, uh, on these tool calls and simplifying the tool calls, which is a very important part. Instead of having a workflow where the master prompt can break into three different branches and then go into four different branches, there's really just a few simple tool calls, uh, including Grep instead of RAG.

And, uh, yeah, and that's kind of what it's trained on. So, uh, these are very optimized tool calling models. So this is, uh, the Zen of Python, if, if you guys are familiar. You do import this in Python.

This is I love this philosophy when it comes to building systems. And I think it's really apt for how Claude Code was built. So really just simple is better than complex. Complex is better than complicated. Flat is better than nested.

This is, this is all you need to this is the whole talk. This is all you need to know about how Claude Code works and why it works specifically. That just in eng we're going back to engineering principles such that simple design is better design.

Uh, I think this is true whether you're building a, uh, database schema, uh, but this is also true when you're building these auten autonomous coding agents. So let's I'm going to now kind of break down all the specific parts of this coding agent and, uh, why I think they're interesting.

So the first is the constitution. Now, a lot of this stuff we kind of take for granted, even though they started doing it a month or two ago or maybe three or four months ago. So this is the Claude MD, Codex or others use agents MD.

The interesting thing, I think I assume most of you know what it is. Uh, it's, again, it's where you put the instructions for your library. But the interesting thing about this is it's basically the team saying, "We don't need to over-engineer a system where the model first researches the repo."

And Cursor, uh, like Cursor 1.0, as you know, makes, uh, vector DB locally to understand the repo and kind of does all this research. They're just saying, "Ah, just put a markdown file. Let the user change stuff when they need.

Let the agent change stuff when they need." Very simple. And kind of goes back to prompt engineering, which I'm a little biased towards because PromptLayer is a prompt engineering platform. But, uh, everything's prompt engineering at the end of the day or context engineering.

Everything is how do you, uh, how do you adapt these general purpose models for your usage? And the simplest answer is the best one here, I think. So this, this is the core of the system. It's just a simple master loop.

Uh, and, and this is actually kind of revolutionary considering how we used to build agents. Everything in Claude Code and, and all the coding agents today, Codex and, and, and, uh, the new Cursor and Amp and all that, it's just one while loop with tool calls, just running the master while loop, calling the tools, and going back to the master while loop.

This is basically four lines of what it's called. I think they call it N0 internally, uh, at least based on my research. But while there are tool calls, run the tool, give the tool results to the model, and do it again until there's no tool calls and then ask the user what to do.

The first time I did this, uh, the first time I used tool calls, it was very shocking to me that the models are so good at just knowing when to keep calling the tool and knowing when to fix their mistake.

And I think that's one of the most interesting things about LLMs. Just they're really good at fixing mistakes and being flexible. And the more just going back, the more you lean on the model to explore and, uh, figure it out, the better and more robust your system is going to be when it comes to better models.

So, so these are the core tools, uh, we have in Claude Code today. And to be honest, these change every day. You know, they're doing new releases every few days, but these are the core ones that I found most interesting to talk about.

Uh, there could be 15 tomorrow. There could be down to five tomorrow. But this is what I find interesting. So first of all, read. Uh, yeah, they could just do a cat. Uh, but what's interesting is read is we have token limits.

So if you've used Claude Code a lot, you've seen that sometimes it'll say this file's too big or something like that. That's why it's worth building this read tool. Grep, glob. Uh, this one's very interesting too because it goes against a lot of the wisdom at the time of using RAG and using vectors.

And I'm not saying RAG has no place, by the way, either. But in these general purpose agents, Grep is good. And, and, and Grep is, uh, how users would do it. And I think that's actually a high-level point here.

As, as you as I'm talking about these tools, remember, these are all human tasks. They're not we're not making up a brand new tool for the model to use. We're kind of just mimicking the human actions and what you and I would do if we were at a terminal trying to fix a problem.

Edit, edit makes sense. I think the interesting thing to note in edit is it's using diffs and it's not rewriting files most of the time. Uh, way faster, way, way, uh, less context used, but also way less, uh, issues.

Uh, if, if I asked you to if I, if I gave you these slides and asked you to review the slides and you read it and had to write down all the slides for me in your new revisions versus if you could just cross out things in the paper, the crossing out is way easier.

Diff is kind of a natural thing to prevent mistakes. Bash, Bash is, uh, Bash is the core thing here, I think. You could probably get rid of all these tools and only have Bash. And the first time I saw this when, when you run something in Claude Code and Claude Code creates a Python file and then runs the Python file, then deletes the Python file, that's, that's the beauty of why this thing works.

So Bash is the most important, I'd say. Web search, web fetch. Uh, the interesting thing about these is they move, move it to a cheaper and faster model. So for example, if you're building a some sort of agent maybe on your platform and you're building an agent and it needs to connect to some endpoints, some list of endpoints, might be worth to bring that into a kind of sub tier as opposed to that master while loop.

That's why this is its own tool. To-dos, uh, we've all seen, seen to-dos. I'll talk about it a little bit more later, but keeping the model on track, steerability, and then tasks. Tasks is very interesting. It's context management.

It's how do we, how do we run this long process, read this whole file without cluttering the context? Because the biggest enemy here is when your context is full, the model gets stupid, for lack of better words. So basically, Bash is all you need.

Uh, I think this is the one thing I want to drill down. The amazing thing about there's two amazing things about Bash for coding agents. The first is that it's simple, uh, and it does everything. It's, it's very robust.

But the second thing that's equally important is there's so much training data on it because that's what we use. It's not it's the reason that models are not as good at Rust or less common programming languages just because there's less people doing it.

So it's really the universal adapter. Um, you've thousands of tools. You could do anything. Uh, this is that Python example I gave. I, I, I always find it so cool when it does the Python script thing or creates tests, and I always have to tell it not to.

But it all these shell tools are in it. And this is I mean, I find myself using Claude Code to spin up local environments where normally I'd have like five commands written down on some file somewhere, and then they get out of date.

It's really good at figuring this stuff out and running the stuff you'd want to do. Uh, and it specifically lets the model try things. So, uh, yeah, the other suggestions here and the tool usage, uh, I think there's a little bit of a system prompt, uh, that tells it which to use and when to use which tool over which.

And this changes a lot, but the these are kind of like the edge cases and the corners you find the model getting stuck in. So reading before editing, uh, they actually may make you do that using Grep the tool instead of the Bash.

So if you look at the tool list here, there's a special Grep tool. Uh, there could be a lot of reasons for that. I think security is a big one, uh, and sandboxing, but then also just that token limit thing, running independent operations in parallel, uh, so kind of pushing the model to do that more.

And then also like these trivial things like quoting paths with spaces. It's just the common, common things. I'm sure they're just dog fooding a lot at Anthropic and they find it and they're like, "Allright, we'll throw it in the system prompt."

Okay, so let's talk about to-do lists. Uh, now again, a very common thing, but was not a common thing before. The, the so this is actually, uh, I think a to-do list for from some of my research for this slide deck.

Um, but the really interesting thing about to-do lists is that they're structured, but not structurally enforced. So here are the rules. One task at a time. Uh, mark them completed. This is kind of stuff you would expect. Uh, keep working on the in-progress if there's block blocks or errors and kind of break up the tasks into different instructions.

But the most interesting thing to me is it's not enforced deterministically. It's purely prompt-based. It's purely in the system prompt. It's purely because our models are just good at instruction following now. And this would not have worked a year ago.

This would not have worked two years ago. Um, there's tool descriptions at the top of the system prompt. We're kind of, uh, injecting the to-dos into the system prompt. Uh, there's they're not, but, but it's not enforced in actual code.

And again, uh, maybe there's other agents that take an opposite path. Uh, I just found this pretty interesting that this at least as a user makes a big difference. And it doesn't even seem it seems it was it seems like it was very simple to implement almost a, a weekend project someone did and seemed to work.

Could be wrong about that as well, but, uh, um, so yeah, it, it's literally a function call. Uh, it, it's the first time you ask something, the reasoning exports this to-do block, and I'll show you what the structure is on the next slide.

Uh, there's IDs. There, there's some kind of structured schema and determinism, but

it, it's just injected there. So here's a example of what it could look like. You get a version, you get your ID, uh, a title of the to-do, and then it could actually inject evidence. So this is, uh, seemingly arbitrary blobs of data it could use.

And the IDs are hashes that it could then refer to, title something human readable. But this is, uh, just another way to structure the data. And in the same way that you're going to organize your desk when you work, this is how we're trying to organize the model.

So I think there's, uh, these are kind of the four benefits we're getting. We're forcing it to plan. Uh, we get to resume after crashes. Uh, Claude Code fails. I think UX is a big part of this. As a user, you know how it's going.

It's not just running off in a loop for 40 minutes without any, uh, signal to you. So UX is non-negligible, even though UX might not make it a better coding agent and might make it better for us all to use.

And, uh, the steerability one. So here's two other parts that were under the hood. Async buffer. So they called it H2A. Uh, it's kind of, uh, the I/O process and how to decouple it from reasoning and, and how to manage context in a way that you're not just stuffing everything you're seeing in the terminal and everything back into the model, which again, context is our biggest enemy here.

It's going to make the model stupider. So we need to, uh, be a little bit smart about that and, and how we do compact and how we do summarization. So here you see when it reaches capacity, it kind of drops the middle, summarizes the head and tail.

Um, then we have the, uh, that's the context compressor there. So what is the limit? 92%, it seems like something like that. Uh, and, and how does it, how does it save long-term storage? That's actually another kind of advantage of Bash in my opinion and having a sandbox.

I would even make a prediction here that all your all ChatGPT windows, all Claude windows are going to come with a sandbox in the near future. It's just so much better because you can store that long-term memory. And I do this all the time.

I have, I have Claude Code skills for deep research and stuff like that. And I'm always instructing it save markdown files because the shorter the context, the quicker it is and the smarter it is. So this is what I'm most excited about.

We don't need DAGs like this. We I'll give you I'll give you a real example. Uh, so some users at PromptLayer, uh, different agents like customer support agent, basically everybody was building DAGs like this for the last two, two and a half years.

Uh, and it was crazy. Hundreds of nodes of, okay, this if this user wants a refund, route them to this prompt. If they want this and a lot of, uh, classifying prompts. The advantage of this is you can kind of guarantee there's not going to be hallucinations or guarantee there's not going to be refunds to people who shouldn't be having refunds or kind of that pro it, it solves the prompt injection problem because if you're in a prompt that purely classifies it as X or Y, injecting doesn't really matter, especially if you throw out the context.

Now, we kind of brought back bring back that attack vector, but the but the major benefit is we don't have to deal with this web of engineering, uh, madness. And, uh, it just it's 10x easier to develop these things, 10x more maintainable, and it actually works way better because our models are just good now.

So this is this is kind of a takeaway is rely on the model. Uh, when in doubt, don't, don't try to think through every edge case and think through every if statement. Just rely on the model to explore and figure it out.

And I was actually two days ago, I think, or yesterday, sometime this week, I was doing an experiment on our dashboard to add like trying these browser agents. And I wanted to see if I could add little titles to all our buttons and it would help the agent navigate our website automatically.

And it actually made it worse, surprisingly. Uh, and maybe I could run it again and maybe I did something wrong with this test, but it made the agent navigate PromptLayer worse because it was getting distracted because I was telling it, "You have to click this button, then you have to click this button."

And then it's it didn't know what to do. So it's better to rely on exploration. You have a question?

Yeah. I'll, I'll push back a little bit.

Please.

I'll admit any scaffolding we create today to resolve the idiosyncrasies of limitations will be like that'll be obsolete three to six months. Even if that's the case, they help a little bit today. I don't how do you balance that like wasted engineering to solve a problem we only have for three months?

It's a great question. So just to repeat, uh, the question is basically, uh, what is the trade-off between solving the actual problems we have today and if you're relying on the model that can't do it yet, but it'll be able to do it in three months,right?

Um, it's case by case. It depends what you're building. If you're building a chatbot for a bank, you probably do want to be a little bit more comp be careful. To me, the happy middle ground is to use this agent paradigm of a master while loop and tool calls, but make your tool calls very rigorous.

So I think it's okay to have a tool call that looks like this or looks like half of this. Uh, in the same way that Claude Code uses read as a tool call or Grep as a tool call.

So for the edge cases, throw it in a structured tool that you can then eval and version and stuff like that. And I could talk I'm going to talk a little bit more about that later, but throw it in that structured tool.

But for everything else, uh, for the exploration phase, leave it to the model or throw some system prompt. Uh, so

it's a trade-off and it's very use case dependent, but I think it's a good question. Thank you. So yeah, uh, just back to Claude Code. Uh, we're, we're getting rid of all this stuff. We're saying we don't want ML-based intent detection.

We don't want regex. We don't want the I mean, it uses regex a little bit, but we don't want regex baked into it. We don't want classifiers. And, and there was a long time we actually built a product for PromptLayer.

We never released it because there was only a prototype of using a ML-based like a non-LLM-based classifier in your prompt pipeline instead of LLMs. And a lot of people have a lot of success with it, but it, it feels more and more like it's not going to be that helpful unless cost is a huge concern for you.

And even then cost is the smaller models is going less and less as, uh, kind of financial engineering between all these companies pays for our tokens. Um, so Claude does also this smart thing, I think, with the trigger phases.

You know, you have think, think hard, think harder, andultra think is my favorite. Uh, and this lets us use the reasoning budget, the reasoning token budget as another parameter that the model can adjust. And this is actually the model can adjust this, but this is how we force it to adjust.

And as opposed to you could make a tool call for hard planning and actually there's some coding agents that do this or you can, uh, let the user specify it and then just on the fly change it. So this is this is one of the biggest topics here, sandboxing and permissions.

I'm going to be completely honest. It's the most boring part of this to me because I just run it on YOLO mode half the time. Um, it's, uh, some people on our team actually dropped all their local databases.

So you do have to be careful. Uh, so, uh, you know, we don't YOLO mode with our enterprise customers, obviously, but, uh, I, but, but I think this stuff is it feels like it's going to be solved, but, but we do need to know how it works a little bit.

So there's a big issue of in prompt injection from the internet. If you're connecting this agent that has shell access and you're doing web fetch, that's a pretty big attack vector. Uh, so there's some containerization of that. There's blocking URLs.

You could see Claude Code's pretty annoying about can I fetch from this URL? Can I do this? And it kind of puts it into a sub-agent. And, uh, yeah, most of the most of the complex code here is in this sandboxing and permission set.

I think there's this whole pipeline to gate Bash commands. So it depending on the prefix is how it goes through the sandboxing environment. And a lot of the other models work differently here. Uh, but this is how Claude Code does it.

I'll explain the other ones later at the end. The next topic, uh, of relevance here is sub-agents. Uh, so this is going back to context management and this, this problem we keep going back to of the longer context, the, the stupider our agent is.

This is a this is an answer to it. So using sub-agents for specific tasks and the key with the sub-agent is it has its own context and it feeds back only the results. And this is how you don't clutter it.

So we got the researcher. These are just four examples. Researcher, docs reader, test runner, code reviewer. In that example I was talking about earlier when I added all the tags to our website to let the agent do it better.

I obviously I use a coding agent to do that. And I said read our docs first and then do it. And it's going to do this in a sub-agent. It's going to feed back the information. And the, the key thing here is the forks of the agent and how we aggregate it back into our main context.

So here's an example. I think this is actually very interesting. I want to call out a thing or two here. So task is what a sub-agent is. We're giving task two things, description and a prompt. The description is what the user's going to see.

So you're going to say task, uh, find default chat context instantiation or something. And then the prompt you're going to give a long string, which is really interesting because now we have the coding agent prompting its own agents.

And I've actually used this paradigm in agents I've built for our product. Uh, if you can you can just have the agent stuff as much information as it wants in this string. And if we're going back to relying on the model, if this task returns an error, now stuff even more information and let it solve the problems.

It's better to be flexible rather than rigid. If I was building this, I would consider switching a string to maybe an object here, uh, depending on what you're building and maybe let it give actually more structured data. Yes.

So I can see this prompt is quite a couple sentences. Is that in the main agent? Is that taking the context in the main agent or is there like, you know, some sort of intermediate step where the sub-agent double reads over, you know, like what the main agent is doing and then, you know, kind of generates a slightly longer prompt or something.

Right. So the question is, does the task just get the prompt here or does it also get your chat history? Is that the question?

Yeah. The question is, is are all of the I have my main agent are is all of this in the system prompt of the main agent to inform how it prompts the sub-agent? No, no. Like it's not in the system prompt.

It's in the whole context. Is the all of this in the context of the main agent?

The task it calls or, or you're saying the structure for the task.

This whole JSON,right? Like or, or.

Yes. So this is a tool call. So the tool call structure of what a task is is in the main agent. Uh, and then these are generated on the fly. Uh, so as you want to run a task, it's generating the description and the prompt.

Task is a tool call. They could be run in parallel and then they're returning the results of it. Hopefully that helps.

Um, so we could go back to the system prompt. So there's some leaks of the Claude Code system prompt. So that's what I'm basing this on. Uh, you can find it online. Um, here are some things I, I noted from it.

Uh, concise outputs, uh, obviously don't give anything too long. No here is or I will just do the do the task the user wants. Uh, kind of pushing it to use tools more, more instead of text explanations. Obviously, I think when we we've all built coding agents and when we do it, it usually says, "Hey, I want to run this SQL."

No, push it to use the tool. Um, matching the existing code, not adding comments. This one does not work for me, but, uh, running commands in parallel extensively and then the to-dos and stuff like that. There's a lot that you can nudge it to do with the system prompts.

But as you see, I think there's a really interesting point to the earlier question you had about where what's the trade-off between DAGs and loops. A lot of these things you could see are feel like they came from someone using it, Claude Code, and saying, "Oh, if only it did this a little less or if it did this a little bit more."

That's where prompting comes in because it's so easy to iterate and it's not you're not it's not a hard requirement, but if only it said here is a little bit more. It's okay to say it sometimes, but allright, skills.

Skills is great. It's, uh, slightly newer. I've I almost got convinced of it only recently. So good. I built these slides with skills. Uh, it's basically, I think, in the context of this talk about architecture, let's think of it as a extendable system prompt.

So in the same way that we don't want to clutter the context, there's a lot of different type of tasks you're going to need to do where you want a lot more context. So this is how we give Claude Code a few options of how it could tap into more information.

Here are some examples. Uh, I use this for I have a skill for docs updates to tell it my writing style and, and my product. So if I want to do a docs update, I say use that skill, load in that skill.

Uh, editing Microsoft Office, uh, Microsoft Microsoft Word and Excel. Um, I, I don't use this, but I've seen a lot of people using it. It kind of like decompiles the file. It's really cool. Uh, but it lets Claude Code do this.

Design style guide. This is a common one. Deep research. I the other day I threw in a like article or GitHub, uh, repo on how deep research works and I said, "Rebuild this as a Claude Code skill." Works so well.

It's amazing. So unified diffing. I think this is worth its own slide. Uh, it's very obvious. Probably not too much we need to talk about here, but it makes this so much better and it makes the token limit shorter.

It makes it faster. It makes it less prone to mistakes. Like I gave with that example when you rewrite an essay versus marking it with a red line. It's just better. I highly recommend using diffing in any agents you're doing.

Unified diff is a standard. When I looked into a lot of these coding agents, some actually built their own kind of standard, uh, and like with slight variations on unified diff because you don't always need the line numbers and but unified diff works.

You got a question?

To go back to skills,

I are I don't know if anyone's seen the Claude the Claude Code warranty and yellow text if your Claude MD gets like greater than 40K characters. And so I was like, "Okay, I'm fucking up. Let me break this down into skills."

So I spent some time and then Claude ignored all of my skills and so I put them somewhere else. So what am I I don't know. Skills feel globally misunderstood or like not I don't know. I'm missing something.

Help me understand.

Yeah. So the the question was on okay. So Claude Code system Claude MD it tells you when it's too long. So, uh, you move it into skills and then it's not recognizing the skills and not picking it up when it's needed.

Yeah. Take that up with the Anthropic team I'd say. Uh, but that's also a good example of maybe the system prompt.

That was the intention like skills you need to invoke them and like the agent itself shouldn't like just call them all the time.

Right. It does give a description of each skill to the model or it should. Uh, tell it, "Okay, here's like a one-liner about each skill." So theoretically in a perfect world it would pick up all the skills all the time, but you'reright.

I generally have to call the skill myself manually. I but I think this is a good tie back into when is prompting theright solution or when is the DAG theright solution or maybe this is a model training problem.

Maybe they need to do a little bit more in post-training of getting the model to call the skills is almost like calling a tool call. You have to know when to call it. So maybe this is just a a functionality that's not that good yet, but I think the paradigm is very interesting, but it's not perfect as we're learning.

So diffing we just talked about. What's next? So this is more opinion-based, but where I see these things going and where the next kind of innovations might likely be. So

I, I there's two schools of thoughts here. A lot of people think we're going to have one master loop with hundreds of tool calls and just tool calling is going to get much better. That's highly likely. Uh, I take the alternate view, which I think we need to reduce the tool calls as much as possible and just go back to just Bash and maybe even put scripts in the local directory.

I think I am on the proponent of one mega tool call instead of a lot of tool calls. Maybe not actually one. I actually think that slide I showed you before is probably a good list, but a lot of people think we need hundreds of tool calls.

I just don't think it's going there. Adaptive budgets, uh, adjusting reasoning. We do this a little bit, uh, the thinking andultra think and stuff like that, but I I think reasoning models as a tool makes a lot of sense as a paradigm.

Can you use I think a lot of us would make a trade-off of a 20 times quicker model with slightly stupider results and being able to call a tool call for a very good model. I think that's a trade-off we'd make in a lot of cases.

Maybe not our planner. Maybe we go to the planner first with GPT-5.1 Codex or Opus or whatever if when the new Opus comes out. Uh, but I think I think there's a lot of, uh, mixing matching we can do and that's I think the next frontier.

And I think the last frontier, I think there's a lot we can learn from to-do lists and and new first-class paradigms we can build. Skills is another example of a first-class paradigm we can kind of try to build into it.

Maybe it doesn't work perfectly. Uh, but I think there's a I think there's a lot of new discoveries to be made there in my opinion. Do I have them? I don't know. Uh, so now I I want to for the for the latter part of this talk, I want to talk about the other frontier agents and the other philosophies they've designed, philosophies they've chosen.

And we all have the benefit. We can mix and match. When we're building our agent, we can do whatever we want and learn from the best. And the frontier labs are very good at this. So, uh, something I like to go back to a lot.

I call it the AI therapist problem. Maybe maybe there's a better name to give it. Uh, but I believe there's a lot of problems. The most interesting AI problems around there isn't a global maximum. Meaning, allright, we're in New York City.

If I need to see a therapist, there's six on every block here. There's no global answer for what the best therapist is. There's different strategies. There's a therapist that does meditation or CBT or maybe one that gives you ayahuasca.

And and these are just kind of like different strategies for the same goal. In the same way that if you're building an AI therapist, there isn't a global maxima. This is kind of my anti-AGI take, but this is also the take to say that when you're building these applications, taste comes into it a lot and design architecture matters a lot.

You can have five different coding agents that are all amazing. Nobody knows which one today. Nobody knows which one's best, to be honest. I don't think Anthropic knows. I don't think OpenAI knows. I don't think Source Graph knows.

Nobody knows whose has the best, but some are better at some things. I personally like Claude Code for I said like running my local environment or using Git or using these kind of like human actions that require back and forth, but I go to Codex for the hard problems or I go to Composer from Cursor because it's faster.

And there's a lot basically all this to say there's value in having different philosophies here and I don't think there's going to be one winner to this. I think there's going to be different winners for different use cases.

And and this is not just coding agents, by the way. This is all AI products. This is this is kind of why our whole company focuses on domain experts and bringing in the PM and the the the subject matter expert into it because that's how you build defensibility.

So here are the perspectives the way I see it. This is not a complete list of coding agents, but these are the ones that I think are the most interesting. Claude Code, I think I think to me it wins in user friendliness and simplicity.

Uh, like I said, if I'm doing something that requires a lot of applications at Git, Git's just the best example. If I want to make a PR, I'm going to Claude Code. Codex, uh, context, it's really good at context management.

Uh, it feels powerful. Do I have the evidence to show you that it's more powerful? Probably not, but, uh, it feels that way to me. And the market feel there's a whole another conversation here to say the market knows best and what people talk about knows best, but I don't know if they know either.

Cursor IDE is kind of that perspective model agnostic. It's faster. Factory, uh, makes Droid, uh, great team. They were here too. Uh, they have multiple they they really specialize these Droid sub-agents they have. So that's kind of their edge and that's maybe a DAG conversation too or maybe a model training.

Uh, cognition. Uh, so Devin, uh, kind of this end-to-end autonomy, self-reflection, AMP, which I'll talk about more in a second. They have a lot of interesting perspectives and actually I find them very exciting these days. Free, it's model agnostic.

Uh, and there's a lot of UX sugar for users. And I actually I love their design, their their talks at this this conference. They they have very very unique perspectives. So let's start with Codex because it's a popular one.

So it's pretty similar to Claude Code. Uh, same master while loop. Most of these do because that's just the winning architecture. Uh, interestingly, Rust Core. Uh, the cool thing is it's open source. You can actually use Codex to understand how Codex works, which is kind of what I did.

Um, it's a little more event-driven, a little more, uh, work went into concurrent threading here. Uh, kind of submission queues, event outputs, kind of the the thing I was talking about with the IO buffer in Claude Code. I think they do it a little bit differently.

Uh, sandboxing is very different. So theirs is more I mean you can see here macOS seat belt and Linux land. Theirs is more kernel-based. Uh, and then

kind of this it's all under threading and and permissions is how I'd say it's mostly different. And then the real difference is the model to be honest. Uh, so this is a this is actually me using Claude Code to understand how Codex works.

Uh, so you see we have a few explore. I didn't talk about explore, but, uh, it's, uh, it's a it's another sub-agent type as you as I as I mentioned, these go in and out. Uh, but yeah, this is researching Codex with Claude Code.

It's always a fun thing to do. So let's talk about AMP. So this is Source Graphs coding agent. I it has a free tier. That's just a cool perspective in my opinion. Uh, they leverage kind of these excess tokens, uh, from providers and they give ads.

We actually have an ad on them. I think it's a cool I'm pro ad. A lot of people are anti-ad. I think it's one of my hot takes, but I like it. They don't have a model selector. This is very interesting too.

This is its own perspective. Uh, it actually helps them move faster because you're you have less of an exact expectation of what the output is because you know they might be switching models here and there. So that changes how they develop.

And then, uh, I think their vision is pretty interesting. Uh, their vision is how do we build not just the best agent, but how do we build the agent that works with the most agent-friendly environments? And actually Factory gave a talk similar to this as well.

But how do how do you build a hermetically sealed, uh, a like coding repo that the agent can run tests on? How do you build the feedback loop? Because that's kind of the holy grail. That's how we build an autonomous agent.

And how do we, uh, I'd love to see the front-end version of this. How do we let it look at its own design and make it better and go back and forth? And this is kind of their guiding philosophy and you could boil down to the agent perspective as I've been calling it.

I think they do interesting stuff with context. So we're all familiar with compact. It's the worst. You have to wait 10 minute. I don't know why it takes so long. Uh, and if you're not familiar, it's summarizing your chat window when the context gets too high and giving the summary.

So they have something called handoff, which makes me think of if you any was anyone was a Call of Duty player back in the day, switch weapons. It's faster than reloading. And, uh, that's what handoff is. You're you're just starting a new thread and you're giving it the information it needs for a new thread.

That feels like the winning strategy to me. Could be wrong, but maybe you need both. That's where they're pushing it. And I kind of like that. I they get they give a very fresh perspective. So the second thing is model choice.

This is the reasoning knobs, uh, and their view on it. They have fast, smart, and oracle. So they lean even more heavily into we have different models. We're not telling you what oracle is. They tell you, but we're willing to switch what oracle is, but we're going to use oracle when we have a very hard problem.

So yeah, so that's AMP. Let's go to Cursor's agent. I think Cursor's agent has a very interesting perspective here. First, obviously, it's UI. Uh, UI first, not CLI. I think they might have a CLI. Not entirely sure, but the UI is the interesting part.

It's just so fast, their new model composer. It's distilled. They have they have the data. They actually made, in my opinion, people interested in fine-tuning again. Fine-tuning was almost, uh, we'd never recommend it to our customers, but Composer shows you that you can actually build defensibility based on your data again, which which is, uh, surprising.

But, uh, yeah, Cursor's agent Composer, I've been almost switching completely to it since because it's just so fast. It's almost too fast. Accidentally pushed to master on one of my personal projects. Uh, so you don't you don't want that always.

Uh, but Cursor was just the crowd favorite and and I want to give a lot of, uh, props to their team. They built iteratively. The first version of Cursor was so bad and it was and we all use I used it because it's a VS Code for fork.

I have nothing to lose and it's gotten so good. It's such a good piece of software and it's a great team. And, uh, but I'll say the same thing can be said about OpenAI's Codex models. They're not quite as fast, but they are optimized for these coding agents and they are distilled.

And I could see OpenAI coming out with a really fast model here because they also have the data. So here's a picture. Um, I think you could this is a picture they put on their blog and you could see what their perspective is on coding agents here just based on the fact that they show you the three models they're running.

So they're offering Composer, but they're letting you use the state-of-the-art because they know that maybe GPT-5.1 is better at planning or here it's five, but now we have 5.1. So here begs the big question, which one should we all use?

Which architecture is best? What should we do? And, uh, my opinion here is that benchmarks are pretty useless. Benchmarks have become marketing for a lot of these model providers. Every model beats the benchmarks. I don't know how that happens, but I think there's there's world where evals matter here.

And the question is what you can eval. The question is how this whole simpl simple while loop architecture that I've been kind of trying to push based on my understanding of it actually makes it harder to eval because if we're relying more on model flexibility, how do you test it?

You could run an integration test, kind of this end-to-end test and just say, does it fix the problem? That's one way to do it. You could break it up. You could kind of do point and time snapshots and say, "Hey, I'm going to give a context to my chatbot from like a half-finished conversation where I know it should be running a specific tool call."

I could run those. Uh, I or I could maybe just run a back test and say, how how often does it change the tools? I think there's also another concept here that's starting to be developed called agent smell, or at least I'm calling it agent smell.

So run an agent and see how many times does it call a tool call? How many times does it retry? How long does it take? And these are all surface-level metrics, but it's really good for sanity checking. And these things are hard to eval.

There's a lot that goes into it. I'll show you an example of what I did, uh, just to kind of dive into it. But but on that subject, maybe I'll just say one more thing. I would break it down.

Me my mental model is you could do an end-to-end test, you can do a point-in-time test, or what I most often recommend is just do a back test. Start with back test. Start capturing historical data and then just rerun it.

So yeah, let me give you, uh, this example. So basically what I have here, so this is a screenshot of PromptLayer. This is our our evals product is also just a batch runner. So you could kind of just run a bunch of columns through a prompt.

But in this case, I'm running it through not a prompt, but Claude Code. So I just have like a head list Claude Code and I'm taking all these providers and I just my head list Claude Code says I think I have it on the next slide.

Search the web for the model provider. It's given to you in a file of variables. Find the most recent and largest model released and then return the name. So I don't know what it's doing. It's doing web search.

I'm not even caring about that. This is an end-to-end test. This is how we kind of try doing Claude Code. And I actually think there's a lot about putting Claude Code into your workflows and those type of headless SDKs.

I'll talk about that, I think, next slide. But kind of main takeaway here is you can kind of start to do end-to-end tests. You can look at it from a high level, do a model smell, and then kind of look into the statistics on each row and see how many times it called a tool.

And going back and we we talked about this a lot in this talk, rigorous tools. The tools can be rigorously tested. You can this is how you offload the this is how you offload the determinism to different parts of your model.

It's you test the tools. You you test the shit out of your tools. Look at them like functions. It's an input and an output. If your tool is a sub-agent that runs, then we're in a kind of recursion here because then you have to go back and test the end-to-end thing.

But for your tools, I'll give you this example. If I so the in my coding agents or my agents in general, my autonomous agents, if there's something very specific that I want to output. So in this case, if I have a very specific type of email format or type of blog post that I want to write and I really want it to get my voiceright, I don't want to rely on the model exploration.

I want to actually build a tool that I can rigorously test. So in this case, this is also just a PromptLayer screenshot, but this is a like a workflow I've built. It has an LLM assertion where it says, "Check if the email's good to my standards.

If it's good, it revises it. If it's not good, it adds the parts." So like the header that it missed, then it revises it with the same step. This is obviously a very simple example, but in I we have another version for some of our SEO blog posts that has like 20 different nodes and writes an outline from a deep research and then fixes a conclusion and adds links.

And for the stuff that you have a very specific vision, that's when testing it just gets so much easier because as you can see, obviously testing this sort of workflow has less steps and less flexibility. So this is an eval I made.

I start with just a bunch of sample emails. I run the prompt or actually I run the the agentic workflow here and I'm just adding a bunch of heuristics. So this is a very simple LMS judge. Does it include three parts in it?

So this is what I was testing for like the high jarred email body and the signature. You can get a lot more complicated. You could do a code execution. You can do I don't know. LMS judge is usually the easiest, but now obviously you could see I could keep running this until it's correct on all of them and kind of, uh, see my eval over time.

This is just from this example. I got it to 100, so that was fun. Uh, and then I want to I want to add another future-looking thing. Keep an eye on headless, uh, Claude Code SDK. I know there was a talk about it this morning.

Um, so I don't want to I won't spend too much time on it, but it's amazing. You just give a simple prompt and it's just another part of your pipeline. I use it for I think I have it on the next slide.

I have a GitHub action that updates my docs every day and just reads all the commits we've pushed to our other repos. And we have a lot of commits going and it just runs Claude Code. Claude Code pulls down all the repos, checks what's updated, reads our Claude MD to see if it should even update the docs, then creates a PR.

So I think this unlocks a lot of things and there's a possibility that we're going to start building agents at a higher order of abstraction and just rely on Claude Code and these other agents to do a lot of the harnesses and orchestration.

Are you reviewing those? Yeah,

I it creates a PR. It doesn't, uh, it doesn't merge the PR. So here are my takeaways. Number one, trust in the model. Uh, when in doubt, rely on the model when you're building agents. Number two, simple design wins.

Number one and number two kind of go together here. Number three, Bash is all you need. Go simple with your tools. Don't have 40 tools, have 10 or five tools. For context management matters, this is the boogeyman we're running from all the time in agents at this point.

Maybe there'll be new models in the future that are just so much better at context, but there's always going to be a limit because, ah, you're talking to a human. I forget people's names if I meet too many in one day.

That's context management or my stupidity. I don't know. And number five, different perspectives matter in agents. I think this is the engineering brain doesn't always comprehend this as much as it should, especially in and I'm an engineer, so I'm also talking about myself, but the different perspectives matter such that there's different, uh, ways to solve a problem where there's not one is better than the other.

And you kind of you probably want a mixture of experts agent. I I would love to have mine run Claude Code and Codex and this and give me the output and consider it a team and maybe have them talk to each other in a Slack-based message channel.

I'm waiting for someone to build that. That would be great. But these are my takeaways. Uh, my bonus thing that I'll show you is how I built this slide deck using Claude Code. So, uh, I built a slide dev skill.

So, I I basically told Claude Code to research how slide dev works and how it can and that's kind of just a library that I made this in. I built a deep research skill to research all these agents and how they work.

I built a design skill because I know if a thing looks terrible or looks good, but I'm not a good designer to figure it out. So, these boxes even I was just like, oh, make the box a little nicer.

Give it an accent color. Uh, so yeah, this is how I built it. But again, thank you for listening. Uh, happy to answer any questions. I'm Jared, founder of PromptLayer. Find me there.

Yes. Thank you. Great talk. Um, so you mentioned, uh, regarding DAGs, basically like let's get rid of them,right? But DAGs kind of enforce this like sequential, uh, execution,right? Of paths. I don't know. Customer service like agent asks the name email,right?

Like in some sort of, um, uh, sequence. Um, so are you saying just write this out, um, like this is now this should be, uh, just written out as a plan for an agent to execute and just trust that the model is going to be calling those tools in that sequence?

Like how do we enforce, uh, the order? Right. So the question was

why do I keep talking about getting rid of DAGs? How else are you supposed to enforce a specific order for solving a problem? So I think there are different types of problems. So the problem of building a general purpose coding agent that we can all use to do our work and even non-technical people can use, there's no specific step to solving that problem, which is why it's better to rely on the model.

If your problem was to build, let's say, a travel itinerary,

it's more of a specific step because you have a deliverable that's always the same. So there's a little bit more of a DAG that could matter, but in the research step of traveling, you probably don't want a DAG because every city is going to be different.

So it really depends on the problem you're solving. I would, if I wanted to make an agent for a travel itinerary, I'd probably have my tool call would one of my tool calls be a DAG of creating the output file because I want the output to look the same or creating the plan.

And then in the system prompt, I could say always end with the output, for example. But you need to mix and match. There's, uh, every use case is different. But if you want to make something general purpose, my take is to rely more on the model on simple loops and less on a DAG.

Cool. Any other questions? Yes. Yeah. Building on that point, like do you think we're heading towards a world where most of you're not actually going to call the API through code and that most LM calls are by triggering Claude Code and just writing the files instead?

So the question is, are we going to move away from calling models directly and just call call like headless Claude Code,right? Yeah. Like if I had a like I have a pipeline that does one LM call per document, summarizes it at the end.

You could make a while loop Claude Code that saves a file every time. You never call the API besides using Claude Code in in a while loop. Potentially. Uh, I'll give you the pro and the con there. Yeah.

The pro is it's easier to develop and we can kind of rely on the frontier. I mean, if you think about it, a reasoning model is just that. The reasoning models didn't always exist. We just had normal LM model and then, oh, now we have O1 and reasoning models.

All that is, I mean, it's a little more complicated in this, but it's basically just a while loop on OpenAI server that keeps running the context and then eventually gives you the output. In the same way that Claude Code SDK is a while loop with a bunch of more things.

So I could totally see a lot of builders only touching these agentic endpoints, maybe even seeing a model provider release a model as an agentic endpoint. But for a lot of tasks, you're going to want a little bit more control and they're probab and probably you'd still want to go as close to metal as possible.

Having said that, there's there's a lot of people who still wanted completions models and that never happened and nobody really talks about that anymore. So it's very likely that everything just becomes this SDK. But I don't have a crystal ball, but those are those are how I I would think about it.

Yes. Thanks for the talk. Um, I always said the simpler the better, but um, what's your thoughts about test-driven development, uh, spec-driven development in AI? Have you tried it? What do you think about it? For building agents or for getting work done?

For coding. Okay. So the question on spec-driven development, test-driven development for coding with agents.

When in doubt, go back to good engineering practices is what I would say. So if you and there's there's whole engineering debates on if test-driven development is theright way and some people swear by it and some people don't.

So I don't think there's an answer. I think coding agents clearly test-driven development makes it easier. I think as I was showing you that's Amp's Source Graph's whole philosophy that if you can build good tests and Factory I think thinks this as well.

If you could build good tests, your coding agent can work much better. So it makes sense to me when I'm working personally, I rely pretty heavily on the planning phase and the spec-driven development phase. And I think the simpler tasks are pretty easy for the model.

But if I'm doing a very simple edit, I'll skip that step. So no one-size-fits-all, but return to the engineering principles that you believe when in doubt, I'd say. Yes. So early early you talked about about system prompt links.

Uh, is it not possible to just look at the uh at the downloaded JavaScript bundle or they have a special endpoint that has prompts behind the endpoint? Yeah. Uh, I think I think they hide it. I think they hide it.

There was a there was actually an interesting article someone because Codex is open source, they before OpenAI released the Codex model that it was using, they were able to hack together the open source Codex to give a custom prompt to the model and be able to use the model without it.

So yeah, you can dive into it, but generally it's tried to be hidden and also laziness of someone posted it. So there you go. That's the work. But someone had to have found it,right? Like is the system prompt somewhere on your machine?

I actually don't know that answer. Do you know that answer? Yeah. Yes. It's on your machine. Nico says it's on your machine. So there we go. So maybe the prompt I was looking at is a little bit old and I have to update it.

But the but uh the question was does uh is the prompt hidden on their servers or can you find it if you are so determined? And the answer seems to be yes.

Any other questions? Yes. Is this the last one? Is this the last question? Can be. Um, you talk about PromptLayer and how can people help you? Yes, that's a good one. I forgot about that. Thank you.

Um, so yeah, my one we're hiring. Uh, so if you're looking for coding jobs at a very fun and fast-moving team in New York, you could reach out to me on X or email jared@promptlayer.com. We're based in New York.

We are uh yeah, we're we're a platform for building and testing AI products for prompt management, audibility, governance, all that fun stuff, but also logging and evals. And those screenshots I showed you came from PromptLayer. So if you're building an AI application and you're building it with a team, you should probably try PromptLayer.

It'll make your life easier. Uh, especially the bigger your team is, the more you want to collaborate, the more you want to collaborate with PMs and non-technical users and or if you're just technical users, it's a great tool.

It'll make your life better. Highly recommend it. promptlayer.com and it's easy to do. And that was my show. Thank you for listening.