Is the Agent Harness Dead? — Jediah Katz, SpaceXAI

AI Engineer · 18 min · 178 sentences · from YouTube's caption track

Each timecode opens YouTube at the start of that sentence. Line anchors (#s42) are the cue ids in the WebVTT, and every line carries its start and end seconds. All transcripts has every talk, and the whole corpus as one file.

00:01[music]
00:12My name is Jedidiah and I joined Cursor early in 2025 as the longest tenured engineer working on our agent harness and recently I came along to SpaceX as part of the acquisition of Cursor.
00:30So, today I want to talk to you all about how the agent harness has evolved to where we are today and where I think things are going.
00:39And there's a motivating question that I have for us to answer today, which is is the agent harness dead?
00:49So, let's get started.
00:54[clears throat]
00:55Here's a more natural and flowing version of that script which reads like a cohesive conference talk rather than a series of short points.
01:02Oops.
01:03Just making sure everyone's still got their soft detectors on before lunch.
01:07Um no, for real.
01:09So, the story starts in early 2023 when prompt engineer was the hottest title out there.
01:17By mid-2025, that had evolved to context engineer, someone who programmatically puts the best context for the model together on the fly.
01:27And then only at the beginning of this year did the industry land on the term agent harness and harness engineering to describe work that my team and I have been doing for a while.
01:40So, what's in an agent harness?
01:43The harness is like the eyes and the hands for the model.
01:48It wraps around the model, steers it, and connects it to the world.
01:53Users connect to our harness through apps and then the harness connects the model to computers and also to third-party services like MCPs.
02:02I like to divide the harness into two layers, the runtime and the context and orchestration layers.
02:10The runtime is kind of like the foundation of the harness.
02:14It controls where the core agent loop runs and how state is stored so that work can resume.
02:21Building this foundation is a lot like traditional software engineering.
02:26And then, the context and orchestration layers are kind of like the wild west.
02:31This is what a lot of people refer to when they talk about the harness.
02:36These parts of the harness control which models are called when and what context they see.
02:44That includes things like prompts and tools, of course, but also memory compaction and a bunch of tricks you might not know that we do to make the agent behave in the way that we want.
02:55So, this is the area that I primarily work in.
02:59In the new year, we kind of hit peak harness.
03:03Cursor announced that we had created a browser entirely from scratch uh using a new long-running harness called Grind Swarm.
03:12This was a multimodal harness that orchestrated a lot of a lot of agents and a lot of machines and it was very expensive to run.
03:21Uh it actually took down Cursor maybe once or twice to to get this demo.
03:26Um now, like anything in the world of agents, it only took a few months of peak harness for people online to start saying that the harness was dead.
03:40Tariq from Anthropic talked about how they had killed much of the system prompt from Claude Code.
03:46Garry Tan from YC thought that the harness would fade away and only skills would matter.
03:54Even myself and my colleague Stefan, we wrote about how we had removed all these guardrails that we had in cursor to patch over gaps in the model's intelligence.
04:04We used to have things like a small model just to apply code edits to a file because the frontier models weren't good at this.
04:11We also had all sorts of coding specific instructions like how to explore the codebase and how to check your work.
04:19So, if all these people are saying the harness is dead, they were saying that smarter models are what killed it.
04:28And there's a lot of evidence for that.
04:31I want you all to think about a very simple harness where the model only has access to the shell and to all of my secrets.
04:39Please don't screenshot this slide by the way.
04:40This one's off limits.
04:43So, who here thinks that this really simple harness would do just as well on coding evals as a more sophisticated one?
04:53Maybe some hands who thinks it would?
04:54Who thinks so Who thinks it would if you Raise your hand if you think it would.
04:58A few hands, but most of you know.
05:00So, a few recent studies tried to investigate exactly that question.
05:05The first one out of UC Berkeley tried seven different frontier models on a series of harnesses including Pi, Codex, and Cloud Code.
05:14And what they found, particularly on the newer smarter models, is that the choice of harness did not actually make a difference significantly in the score on SweetBench light.
05:26And then another another study from from Apple and EPFL found similarly that a dedicated MLE harness did not make a big difference on MLEBench.
05:36So, what is going on here?
05:39Is it Is it actually true that the harness is dead?
05:40Are they right?
05:42Here's what I think that the evidence shows.
05:45A fancy harness in this day and age, no longer adds coding intelligence.
05:49I think that is true.
05:51I know that's surprising.
05:54But, these things that we used to do in the harness to patch over gaps in the raw coding muscle memory of the model, like teaching it how to explore the code base,
06:03uh how to work persistently for very long time with things like slash goal, uh how to run tests to check your work.
06:10The model has natively absorbed these skills.
06:16So, why is this happening?
06:20It turns out the harness is actually making itself obsolete through what I call the data flywheel.
06:27So, in this process, the harness brings harder and harder coding tasks within reach.
06:34At that point, more users can productively attempt those tasks with agents.
06:38That means more coding data is shared with the labs.
06:41And eventually the models learn natively how to do these skills as well.
06:47Eventually, the models get so good that the harness can actually delete the things it needed to do to make those tasks work in the first place.
06:57It's a very interesting process.
07:00So, that's how you get people saying that the harness is dead.
07:04And if the models are good enough to code well in any harness, then why should you care?
07:08Does it really matter which harness you use?
07:12Is the harness really dead?
07:14I say no, and there's two reasons why.
07:18First of all, and I'm going to go into both these points, the harness actually opens the new frontier of tasks for the model to solve.
07:26And then, second, there are some tasks or some parts of the harness, especially outside the context layer, which the model cannot absorb.
07:34So, to first talk about how the harness unlocks the next frontier, I want to talk about Grokbot.
07:41Who here has tried Grockbot?
07:42Can I see some hands?
07:44Awesome.
07:44Let's go.
07:45That's great.
07:47Um so, for those of you who haven't, Grockbot is our autonomous colleague who gets work done for you.
07:52Not just coding, but all kinds of knowledge work.
07:56And I really recommend you try it.
07:58I'm not going to sell too much, but it's a groundbreaking new product.
08:00It's already created a new category.
08:02Let's just talk about the harness, though.
08:05So, Grockbot has a new kind of harness that operates not based off of one-off chats, but long-lived coordinator bots that orchestrate a fleet of worker agents.
08:16You can kind of think of it like hiring a high-powered lawyer who deals with you, you know, directly and owns strategy, but they farm out their tasks to a team of paralegals.
08:29It's like that.
08:32Now, something interesting about the Grockbot harness is I mentioned how in Cursor we found ourselves removing things from the harness, removing prompting, removing guardrails as the models got smarter.
08:43But, curiously in Grockbot, we've actually found ourselves putting back a lot of these things, adding new prompting, adding new guardrails.
08:50And that's because there's a lot of new tasks in this new way of working, in the new domains that Grockbot is meant to solve, that the models have not nailed yet.
09:01To give a few examples of those kind of things, just a few things that we're focused on right now when it comes to Grockbot, there's key website flows like making slide decks and booking travel,
09:11communication.
09:12An autonomous colleague needs to keep you up to date on what it's been doing without overwhelming you.
09:17Acting versus asking is another big one.
09:19Grockbot can do a lot of things which are kind of risky in some sense, like emailing your colleagues.
09:24You want us to be able to do that, but you only want us to do it in the way that you intended.
09:28And then, of course, learning over time the way that you work through memories is another big focus.
09:36So, those of you who might have pushed the boundaries with coding agents might have noticed that none of these tasks are new.
09:43But, for GrokBot, they're much more important because everybody is needing them in some way.
09:49And right now in GrokBot, it is the harness that is making these tasks feasible.
09:54Of course, the model is a big part of that, but the harness is helping bring them into its own reach.
09:59So, going back to our data flywheel, GrokBot, that example, lets us fill in the last the last arrow in our cycle, which is when a capability like coding gets commoditized by the model,
10:13it's actually the harness which unlocks a new frontier for the model to then solve and kicks off that flywheel again.
10:22So, that might look like, you know, taking on new tasks outside of the coding domain, um as we just saw with GrokBot, or it could just mean taking on tasks in the software engineering job beyond
10:35pure coding.
10:36I'm sure some of you here are, you know, pretty senior in your software careers, and you might know the more senior you get, the less coding you're actually doing.
10:42So, there's a lot of other things in the job that matter.
10:44Um this is the core idea that I would like you to understand, this flywheel and how the model and the harness together drive it.
10:53One more example of how the harness can unlock new frontiers is some of you may have heard of this idea of the software factory, where you have multiple agents coordinating together
11:05to solve the full software development life cycle.
11:09Now, this is far from a solved vision, but there's a lot of industry or there's a lot of innovation happening across the industry at the harness.
11:18To give an example of how this might manifest, you know, internally, we have a lot of channels where someone might report a bug to that channel, and then an automation
11:26immediately auto triages that bug.
11:28Then it might kick off a cloud agent to try to fix the bug, and that agent will use our tooling to show before and after videos to prove that it's fixed.
11:36And then, once that fix is deployed and or is merged in and starts deploying, we have our rollouts product, which kicks off another agent to watch in production for any issues.
11:50So, you can think of all this as almost like a meta harness, which is orchestrating many agents together.
11:56That's how I'd like you to think about this.
12:00So, that's the first reason you should care about the harness because a frontier harness is needed to be at the frontier of agentic tasks.
12:07And then the next reason is there are some parts of the harness that the model can't absorb.
12:12So, let's talk about those.
12:16First of all, there is the infrastructure and runtime needed to make an autonomous colleague reliable.
12:21Over the past year plus, we've gone from agents that work on a typical task for minutes to hours to days and even longer.
12:30And there's just no way that this can work on one person's laptop, which might go to sleep.
12:36It has to be running in the cloud if you really want an autonomous colleague.
12:43And then next, we have trust and security.
12:45Yes, like we saw before, uh a simple harness with just bash and all your, you know, all your API keys in a file can solve coding tasks, but you probably don't want to give all your API keys in a file
12:56to the model.
12:57Um or if you're okay with that, I'm sure that your head of security would have something to say.
13:02Um so, this is obviously really important companies where we want to have permissions that are scoped down to specific agents and specific people, but it's also important to have guardrails for anyone who doesn't want to wake up one morning to a really catastrophic action or to a $10,000 API fee, right?
13:21And then finally, there's efficiency.
13:24If you think back to to two studies I showed, where the choice of harness did not make a significant difference in eval scores, both of those studies actually found
13:33that there was a significant difference in cost per task.
13:38I have personally been focused over the past month on optimizing the efficiency of the Grok Bot harness.
13:43Now, through tweaking variables like uh cache hit rate, improving that, token efficiency, um model mix, we were able to bring down the costs and the speed of Grok Bots significantly
13:55for users without any hit to quality.
13:58Users have really, really been appreciating this.
13:59It's been awesome to see.
14:01Um now, it could be an entire different talk about how we did this, but the key takeaway here is that the more and more that you use agents to do your work, the more it matters, the more valuable it will be for you to have the most efficient harness.
14:18So, by now you might be convinced that the harness still matters.
14:21Then you might be asking, "What's next?"
14:24Here's a few trends that I'm anticipating.
14:27First of all, that flywheel is speeding up, and right now we've entered a new age where the product is once again moving faster than the model.
14:35That means that the harness is going to get more messy before it gets cleaned up.
14:39There's going to be more things shipped.
14:41So, because of this, we're expecting people to um build their harnesses, design them so it's easier to ship and unship new features.
14:50One way we're already seeing that manifest is with both Grok Bot and other similar products, less and less of the harness inner workings are being exposed to the user.
14:58Some of you may have heard of Hyrum's Law, which says that anything that you expose to users, they will come to depend on.
15:05Now, obviously there are tradeoffs to giving users less, you know, less visibility into what's happening, but one advantage is that there's less change management to deal with, and you can iterate much faster.
15:18Now, another way that we've been seeing um this trend manifest is a bigger focus on designing primitives for agents.
15:26Primitives are like building block interfaces that can be composed and extended into powerful features.
15:32When you have primitives, you can add new features without making things more complicated for the agent, and you can remove features without having to do a complicated refactor in your code base.
15:41It's just good engineering.
15:43One example across our agent harnesses of a primitive we've added is called subscriptions.
15:48Agents need a way to wake up in response to events.
15:51Those might be someone posting in Slack, CI going red on a PR that the agent cares about, or just a certain time of day.
16:00Now, rather than having one separate integration, you know, having separate integrations for each of these things, we have one framework and one tool that the model can use or the agent can use to subscribe to an event.
16:12And then another cool thing that we do, I mentioned earlier that we have these deterministic guardrails to patch over bad behaviors in the model.
16:20That might be, for example, if GrokBot doesn't respond to the user fast enough, and we might give it give it a little nudge and say, "Hey, user's waiting on you."
16:28Now, we'd rather not do these things.
16:31They're kind of like hacks, but we would rather fix them now than, you know, wait for the next model checkpoint to to get released.
16:36So, in order to make our code base, you know, be less messy and and have less of a worse time iterating the future, we track all these things, which we call shims centrally,
16:46and every time the model uses one, it gets logged.
16:48So, now we have a bot who watches this dashboard, and whenever a shim stops firing, that model or that agent will put up a PR and say this shim can be cleaned up.
16:58Another thing that I think is next for the harness is it is moving up to a higher level of abstraction.
17:04Yesterday's generation of harnesses were built from composing a bunch of LLM calls.
17:08Now, the new harnesses we're seeing, like GrokBot and the software factory, are using agents as the core building block.
17:15And then finally, I predict that within 6 months, 90% of agent work is going to run off the laptop.
17:22This is already where things are going, and I think there's just no way that, you know, this is this is not necessary in order to have truly autonomous agents who can, you know, really work with us in in the way that we want.
17:36So, is the harness dead?
17:39The parts that made it smarter at coding, yes, that is the flywheel at work.
17:46That is the flywheel at work.
17:47But, the harness, as we saw, is how the agents can unlock the next frontier of work.
17:53And then also the trust, the efficiency, the infrastructure that the harness brings, those things are not going anywhere.
18:00So, yesterday's harness may be dead, but long live the agent harness of tomorrow.
18:07Thank you.
18:08[applause]