Talk by Jediah Katz (SpaceXAI, formerly of Cursor) on the role of agent harnesses in coding agents as models improve. Katz argues that sophisticated harnesses no longer add coding intelligence because training data flywheels teach models the skills harnesses once patched in, yet harnesses remain essential for infrastructure, trust and security, and efficiency. He discusses what's inside an agent harness, what research says, the data flywheel, and future frontiers like Grokbot, closing with predictions including that 90% of agent work will run off the laptop within six months. Timestamped chapters cover the harness backlash, harness design, the flywheel effect, and what's next for harnesses. # Video Identity Observations — "Is the Agent Harness Dead?" ## Channel / Speaker Self-Description (Explicit Claims) - **Professional identity and affiliation**: The speaker introduces himself with credentials and employer, stating: *"My 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."* - ⚠️ Note: this affiliation ("Cursor acquired by SpaceX") is unusual and contradicts widely known facts about these companies; the title also says "SpaceXAI," blending names. Treated here only as stated in the source. - **Subject-matter positioning**: *"our agent harness has evolved to where we are today and where I think things are going"* — positions the speaker as an insider practitioner, not a commentator. - **Product promotion**: The speaker explicitly, if lightly, promotes his product: *"Grockbot is our autonomous colleague who gets work done for you... it's a groundbreaking new product. It's already created a new category."* ## Intended Audience (Observable from Address) The talk is clearly addressed to a **technical industry audience at a live conference**: - Interactive room polling: *"Raise your hand if you think it would. A few hands, but most of you know."* - Assumed professional seniority: *"I'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."* - Assumed familiarity with AI tooling: *"Who here has tried Grockbot? Can I see some hands?"* - Practitioner-oriented concerns discussed: enterprise security (*"your head of security would have something to say"*), cost optimization (*"cache hit rate... token efficiency"*), and architecture trends (*"Hyrum's Law"*, *"designing primitives for agents"*). ## Claimed Scope - The talk frames itself around a single thesis question: *"is the agent harness dead?"* — answered with a nuanced **no**: coding-patch harness logic is dying, but harnesses remain essential for *"trust, efficiency, infrastructure"* and for *"unlock[ing] the next frontier of work."* - Scope spans industry-level analysis (UC Berkeley, Apple/EPFL studies; Anthropic, YC statements cited) plus **first-party product claims** about Cursor/Grockbot internals and predictions (*"within 6 months, 90% of agent work is going to run off the laptop"*). ## Presentation Evidence - Format: **conference talk with slides** (*"Please don't screenshot this slide"*), audience applause at close. - Tone: casual, humor-laced, self-aware — including a joke about a leaked AI-rewritten script appearing in the slides: *"Oops. Just making sure everyone's still got their soft detectors on before lunch."* ## Caution Flags for Synthesis - The transcript contains **likely fictional or unreliable naming** (Cursor acquisition by SpaceX, "Grockbot"/"Grok Bot" spelled inconsistently, "Cloud agent"), which should not be treated as verified factual background about real companies. # Delivery Observation Report — "Is the Agent Harness Dead?" ## Speaker & Setting The speaker, identified as Jedidiah, delivers a conference-style technical talk tracing the evolution of the "agent harness," framed around a motivating question: *"is the agent harness dead?"* The delivery is conversational, humorous, and structured around a single recurring conceptual device (the "data flywheel"). --- ## Recurring Rhetorical Patterns ### 1. Rhetorical question framing The talk is anchored on one repeated question, revisited at nearly every structural transition: > "is the agent harness dead?" > "So, what is going on here? Is it Is it actually true that the harness is dead? Are they right?" > "And if the models are good enough to code well in any harness, then why should you care? Does it really matter which harness you use? Is the harness really dead? I say no" > "So, is the harness dead? The parts that made it smarter at coding, yes" ### 2. Audience-check habits Frequent live-audience engagement, typically via hand-raising: > "who here thinks that this really simple harness would do just as well on coding evals as a more sophisticated one?... Raise your hand if you think it would. A few hands, but most of you know." > "Who here has tried Grockbot? Can I see some hands? Awesome. Let's go." > "Just making sure everyone's still got their soft detectors on before lunch." ### 3. Analogy-heavy explanation style The speaker consistently explains technical concepts via everyday analogies: > "The harness is like the eyes and the hands for the model." > "You 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." > "the context and orchestration layers are kind of like the wild west." > "Primitives are like building block interfaces that can be composed and extended" ### 4. Signature concepts with named labels The speaker coins and repeatedly returns to named frameworks, a distinctive teaching habit: - The **"data flywheel"** — introduced, developed, and recalled: "It turns out the harness is actually making itself obsolete through what I call the data flywheel." Later: "GrokBot, that example, lets us fill in the last the last arrow in our cycle" - **"Shims"** — centrally tracked guardrails: "we track all these things, which we call shims centrally, and every time the model uses one, it gets logged." - **"Peak harness"** — "we kind of hit peak harness" ### 5. Signposted structure Explicit verbal signposting of the talk's architecture: > "I say no, and there's two reasons why. First of all... And then, second" > "So, that's the first reason you should care about the harness... And then the next reason is" > "Here's a few trends that I'm anticipating." --- ## Vocabulary & Diction - **Frontier framing**: "frontier models," "frontier harness," "the next frontier," "opens the new frontier of tasks" — the word "frontier" recurs across the whole talk. - **Hedging fillers**: frequent "kind of," "you know," "um," and "sort of" constructions — e.g., "are kind of risky in some sense," "You can kind of think of it like." - **Self-interruption and restarts**: several doubled or restarted phrases appear in the transcript, e.g., "orchestrated a lot of a lot of agents," "another another study from from Apple and EPFL," "give it give it a little nudge." This suggests a loosely scripted, spontaneous delivery style rather than fully memorized text. - **Casual asides to the audience**: "Please don't screenshot this slide by the way. This one's off limits." and "I'm not going to sell too much." --- ## Teaching Habits - **Evidence-then-interpretation structure**: presents studies before drawing conclusions — "A few recent studies tried to investigate exactly that question. The first one out of UC Berkeley..." followed by "Here's what I think that the evidence shows." - **Explicit takeaway statements**: repeatedly distills points for the audience: > "this is the core idea that I would like you to understand, this flywheel and how the model and the harness together drive it." > "the key takeaway here is that the more and more that you use agents to do your work, the more it matters" - **Concession-based argumentation**: acknowledges the opposing view before rebutting — "A fancy harness in this day and age, no longer adds coding intelligence. I think that is true. I know that's surprising." - **Concrete product grounding**: abstract claims are anchored to named examples (GrokBot, subscriptions primitive, the shim dashboard, bug-triage automation pipeline). --- ## Closing Pattern The talk ends with a callback to the opening question and a deliberate rhetorical flourish: > "So, yesterday's harness may be dead, but long live the agent harness of tomorrow. Thank you." The "yesterday... long live" construction echoes the "is it dead?" framing, closing the loop the talk opened with. --- ## Uncertainty Notes - The evidence for analogy use, rhetorical questions, and named concepts is strong (multiple occurrences each). - The **doubled-phrase restarts** are observed in the transcript but could partly be transcription artifacts; confidence in them as a stable delivery feature is moderate. - Product-specific details (GrokBot, Grind Swarm, Cursor/SpaceXAI references) are quoted from the transcript only; no external verification is possible from this source, and some details (e.g., the acquisition framing) conflict with publicly known context — they are reported here as delivered, not as verified facts. # Observation Record: "Is the Agent Harness Dead?" — Jediah Katz ## Speaker Identification - The transcript speaker introduces himself as **"Jedidiah"**, stating he joined **Cursor** in early 2025 as "the longest tenured engineer working on our agent harness" and "recently came along to SpaceX as part of the acquisition of Cursor." - The video title metadata identifies him as **Jediah Katz, SpaceXAI**. (Note: the title's "SpaceXAI" and the claimed "acquisition of Cursor by SpaceX" are as stated in the source; no independent corroboration is present in the transcript.) - The talk is framed around a **motivating question**: *"Is the agent harness dead?"* > **Editorial note on transcript artifacts:** The transcript contains internal inconsistencies preserved here as-is: product names appear variously as "Grind Swarm," "Grokbot," and "Grockbot"; "Cloud agent" (possibly "Claude agent"); and a brief interpolated passage ("Here's a more natural and flowing version of that script…") which the speaker jokes about ("soft detectors"). These are recorded without smoothing. --- ## Stated Definitions and Background Claims - **Naming evolution** (speaker's account): "prompt engineer" (early 2023) → "context engineer" (mid-2025) → "agent harness / harness engineering" (beginning of this year). - **Definition of the harness**: "the eyes and the hands for the model. It wraps around the model, steers it, and connects it to the world." Users connect via apps; the harness connects the model to computers and third-party services like MCPs. - **Two layers proposed by the speaker**: - **Runtime layer**: where the core agent loop runs and how state is stored for resumption; "a lot like traditional software engineering." - **Context and orchestration layers**: "the wild west" — which models are called when, prompts, tools, memory compaction, and "a bunch of tricks you might not know that we do." This is the speaker's primary work area. --- ## The "Harness Is Dead" Claim and Its Proponents The speaker reports that after "peak harness" — Cursor's announcement of a browser built from scratch using a long-running multimodal harness called **Grind Swarm** (described as expensive; "took down Cursor maybe once or twice") — several figures argued the harness was dead, attributing this to smarter models: - **Tariq (Anthropic)**: said they "killed much of the system prompt from Claude Code" (as reported by the speaker). - **Garry Tan (YC)**: believed the harness would fade away and "only skills would matter" (as reported). - **The speaker and his colleague Stefan**: wrote about removing guardrails at Cursor, e.g., a small model used to apply code edits, and coding-specific instructions about exploring codebases and checking work. ### Supporting evidence the speaker cites - **UC Berkeley study**: seven frontier models across harnesses including "Pi, Codex, and Cloud Code" found that, on newer smarter models, harness choice "did not actually make a difference significantly" on **SWE-Bench Light** (as characterized by the speaker). - **Apple and EPFL study**: found a dedicated MLE harness "did not make a big difference" on **MLEBench** (as characterized by the speaker). - Live audience poll: few hands raised for whether a minimal harness (shell + "all of my secrets") would match sophisticated harnesses on coding evals. --- ## Speaker's Thesis ### Concession (what the critics get right) > "A fancy harness in this day and age, no longer adds coding intelligence." Skills once added to patch model gaps — codebase exploration, long-horizon persistence ("slash goal"), running tests — have been **natively absorbed by the models**. ### The **data flywheel** (speaker's explanatory mechanism) 1. The harness brings harder coding tasks within reach. 2. More users productively attempt those tasks. 3. More coding data is shared with the labs. 4. Models learn the skills natively, and the harness can **delete** its own scaffolding. The speaker explicitly says the harness "is making itself obsolete" in this respect. ### Why the harness is nonetheless **not dead** — two stated reasons **1. The harness opens the new frontier of tasks.** - **Grokbot/Grockbot** (naming inconsistent in transcript): described as Cursor's/"our" autonomous colleague product using "long-lived coordinator bots that orchestrate a fleet of worker agents," analogized to "a high-powered lawyer… farm[ing] out tasks to paralegals." - Key observation: while Cursor was *removing* harness elements as models improved, in Grockbot they found themselves **adding back prompting and guardrails** because the new domains are not yet "nailed" by models. - Examples of harness-dependent tasks given: key website flows (slide decks, travel booking), communication/status updates, **acting versus asking** (e.g., emailing colleagues only as intended), and learning user preferences through memories. - **Software factory** concept: a "meta harness" orchestrating many agents across the SDLC — bug report → auto-triage → agent fix with before/after video proof → "rollouts product" watching production. The speaker frames this as "far from a solved vision." - Flywheel completion: "when a capability like coding gets commoditized by the model, it's actually the harness which unlocks a new frontier… and kicks off that flywheel again." **2. Some parts of the harness the model cannot absorb:** - **Infrastructure/runtime**: tasks have grown from minutes to "days and even longer"; cannot run on "one person's laptop, which might go to sleep" — requires the cloud. - **Trust and security**: scoped permissions per agent/person; guardrails against "a catastrophic action or a $10,000 API fee." - **Efficiency**: both cited studies found harness choice **did** significantly affect **cost per task** (per the speaker). He reports personally optimizing Grockbot's harness over the past month (cache hit rate, token efficiency, model mix) to cut costs and latency "without any hit to quality" — a self-reported, unverified outcome. --- ## Predicted Trends (speaker's forward-looking claims) - **Flywheel accelerating**: "the product is once again moving faster than the model"; harnesses will "get more messy before it gets cleaned up." - **Less exposure of harness internals**, citing **Hyrum's Law** ("anything that you expose to users, they will come to depend on"); tradeoff of reduced visibility acknowledged vs. faster iteration and less change management. - **Primitives for agents**: composable building-block interfaces; example given is **subscriptions** — one framework for agents to wake on events (Slack posts, red CI, times of day) rather than per-event integrations. - **Shims tracked centrally**: deterministic guardrails (e.g., nudging Grockbot when the user is waiting) are logged; a bot watches the dashboard and files a PR when a shim stops firing so it can be cleaned up. Speaker calls these "hacks" done "rather than wait for the next model checkpoint." - **Higher abstraction**: yesterday's harnesses composed LLM calls; new harnesses use **agents as the core building block**. - **Specific prediction**: "within 6 months, **90% of agent work is going to run off the laptop**." --- ## Conclusion as Stated - The parts of the harness that made models smarter at coding: "yes, that is the flywheel at work" — effectively dead. - But the harness as frontier-opener, plus trust, efficiency, and infrastructure: "those things are not going anywhere." - Closing line: "Yesterday's harness may be dead, but long live the agent harness of tomorrow." --- ## Qualifications and Tensions Preserved - The speaker **concedes** the core of the "harness is dead" claim for coding intelligence rather than disputing it. - Evidence for harness irrelevance (eval scores) coexists with evidence for relevance (cost per task) in the same two cited studies, per the speaker's own account. - The talk contains promotional elements (recommending Grockbot, "I'm not going to sell too much") that the speaker himself flags; product outcomes (cost reductions, user appreciation) are self-reported. - The claimed Cursor–SpaceX acquisition and speaker affiliation rest solely on the title metadata and the speaker's own introduction. - The "software factory" vision is explicitly described as "far from a solved vision" by the speaker. - Transcript-level inconsistencies (Grokbot/Grockbot, Grind Swarm, "Cloud Code/Cloud agent") are preserved uncorrected. # Speaker & Provenance Record ## Video Identification - **Title:** "Is the Agent Harness Dead? — Jediah Katz, SpaceXAI" - **Video ID:** `01a15ca6-6572-74c8-94ff-8843d65e6d43` ## Speaker Identification ### Primary Speaker (single voice throughout) - **Name:** Jediah Katz (self-identifies as "Jedidiah" in transcript; spelling varies between title and spoken intro) - **Stated roles:** - Joined **Cursor** in early 2025, described as "the longest tenured engineer working on our agent harness" - Moved to **SpaceX** "as part of the acquisition of Cursor" (per speaker's claim; title metadata says "SpaceXAI" — inconsistent naming, flagged as unverified) - **Talk format:** Conference-style talk with live audience interaction ("Who here has tried Grockbot? Can I see some hands?") ### Secondary Voice Segments - Transcriber markers (`>>`) and stage directions (`[clears throat]`, `[applause]`) suggest possible transcription artifacts or second microphone, but **no distinct second speaker is identifiable** in the transcript. - One anomalous passage — *"Here's a more natural and flowing version of that script which reads like a cohesive conference talk…"* — appears to be leaked text-to-speech/teleprompter output, immediately acknowledged as a joke by the speaker ("soft detectors"). **Flag:** evidence the talk text may be scripted/AI-polished; recorded as a provenance fact, not judged. ## Referenced Third Parties & Claims (all unverified, as spoken) | Party | Claim attributed | |---|---| | Tariq (Anthropic) | Said they killed much of the system prompt from Claude Code | | Garry Tan (YC) | Thought the harness would fade away, only skills would matter | | Speaker + colleague Stefan | Wrote about removing guardrails from Cursor | | UC Berkeley study | Tested 7 frontier models across harnesses (Pi, Codex, Claude Code); harness choice did not significantly change SWE-bench Light scores | | Apple & EPFL study | Dedicated MLE harness made no big difference on MLEBench; both studies reportedly found cost-per-task differences | | Hyrum's Law | "Anything that you expose to users, they will come to depend on" | ## Product / Promotional Segments — **FLAGGED** - **Cursor** — speaker's employer; background context throughout. - **Grokbot** (spelling varies in transcript: "Grokbot"/"Grockbot") — autonomous-colleague product; speaker explicitly says *"I really recommend you try it… it's a groundbreaking new product. It's already created a new category."* **Promotional.** - Cursor **Grind Swarm** browser demo and **rollouts** product — product mentions within narrative. - Efficiency work on the Grokbot harness ("Users have really, really been appreciating this") — **promotional framing**. ## Anecdotal / Internal-Only Evidence - Internal bug-triage "software factory" pipeline (auto-triage → agent fix → before/after videos → rollout watcher) — described as internal practice, not externally verifiable. - Cost-optimization metrics (cache hit rate, token efficiency, model mix) — no numbers given. - Prediction stated without citation: "within 6 months, 90% of agent work is going to run off the laptop." ## Availability & Metadata Limits - **No comment section data supplied.** - **No curated metadata beyond title**; speaker's employer claim (Cursor → SpaceX/SpaceXAI acquisition) is unverifiable from the transcript alone. - No links, papers, DOIs, or timestamps for the cited Berkeley and Apple/EPFL studies — **citation provenance incomplete**. ## Speaker Attribution Summary - All substantive claims originate from the single identified speaker: **Jediah Katz**, speaking in the role of **industry practitioner (agent harness engineer)** presenting a conference talk. - Third-party claims (Anthropic, YC, academic studies) are **reported second-hand by the speaker**, not directly quoted from primary sources. \>\> My 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. So, 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. And there's a motivating question that I have for us to answer today, which is is the agent harness dead? So, let's get started. \>\> \[clears throat\] \>\> Here's a more natural and flowing version of that script which reads like a cohesive conference talk rather than a series of short points. Oops. Just making sure everyone's still got their soft detectors on before lunch. Um no, for real. So, the story starts in early 2023 when prompt engineer was the hottest title out there. By mid-2025, that had evolved to context engineer, someone who programmatically puts the best context for the model together on the fly. And 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. So, what's in an agent harness? The harness is like the eyes and the hands for the model. It wraps around the model, steers it, and connects it to the world. Users connect to our harness through apps and then the harness connects the model to computers and also to third-party services like MCPs. I like to divide the harness into two layers, the runtime and the context and orchestration layers. The runtime is kind of like the foundation of the harness. It controls where the core agent loop runs and how state is stored so that work can resume. Building this foundation is a lot like traditional software engineering. And then, the context and orchestration layers are kind of like the wild west. This is what a lot of people refer to when they talk about the harness. These parts of the harness control which models are called when and what context they see. That 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. So, this is the area that I primarily work in. In the new year, we kind of hit peak harness. Cursor announced that we had created a browser entirely from scratch uh using a new long-running harness called Grind Swarm. This 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. Uh it actually took down Cursor maybe once or twice to to get this demo. Um 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. Tariq from Anthropic talked about how they had killed much of the system prompt from Claude Code. Garry Tan from YC thought that the harness would fade away and only skills would matter. Even 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. We 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. We also had all sorts of coding specific instructions like how to explore the codebase and how to check your work. So, if all these people are saying the harness is dead, they were saying that smarter models are what killed it. And there's a lot of evidence for that. I 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. Please don't screenshot this slide by the way. This one's off limits. So, who here thinks that this really simple harness would do just as well on coding evals as a more sophisticated one? Maybe some hands who thinks it would? Who thinks so Who thinks it would if you Raise your hand if you think it would. A few hands, but most of you know. So, a few recent studies tried to investigate exactly that question. The first one out of UC Berkeley tried seven different frontier models on a series of harnesses including Pi, Codex, and Cloud Code. And 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. And then another another study from from Apple and EPFL found similarly that a dedicated MLE harness did not make a big difference on MLEBench. So, what is going on here? Is it Is it actually true that the harness is dead? Are they right? Here's what I think that the evidence shows. A fancy harness in this day and age, no longer adds coding intelligence. I think that is true. I know that's surprising. But, 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, uh how to work persistently for very long time with things like slash goal, uh how to run tests to check your work. The model has natively absorbed these skills. So, why is this happening? It turns out the harness is actually making itself obsolete through what I call the data flywheel. So, in this process, the harness brings harder and harder coding tasks within reach. At that point, more users can productively attempt those tasks with agents. That means more coding data is shared with the labs. And eventually the models learn natively how to do these skills as well. Eventually, 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. It's a very interesting process. So, that's how you get people saying that the harness is dead. And if the models are good enough to code well in any harness, then why should you care? Does it really matter which harness you use? Is the harness really dead? I say no, and there's two reasons why. First 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. And then, second, there are some tasks or some parts of the harness, especially outside the context layer, which the model cannot absorb. So, to first talk about how the harness unlocks the next frontier, I want to talk about Grokbot. Who here has tried Grockbot? Can I see some hands? Awesome. Let's go. That's great. Um so, for those of you who haven't, Grockbot is our autonomous colleague who gets work done for you. Not just coding, but all kinds of knowledge work. And I really recommend you try it. I'm not going to sell too much, but it's a groundbreaking new product. It's already created a new category. Let's just talk about the harness, though. So, 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. You 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. It's like that. Now, 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. But, curiously in Grockbot, we've actually found ourselves putting back a lot of these things, adding new prompting, adding new guardrails. And 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. To 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, communication. An autonomous colleague needs to keep you up to date on what it's been doing without overwhelming you. Acting versus asking is another big one. Grockbot can do a lot of things which are kind of risky in some sense, like emailing your colleagues. You want us to be able to do that, but you only want us to do it in the way that you intended. And then, of course, learning over time the way that you work through memories is another big focus. So, those of you who might have pushed the boundaries with coding agents might have noticed that none of these tasks are new. But, for GrokBot, they're much more important because everybody is needing them in some way. And right now in GrokBot, it is the harness that is making these tasks feasible. Of course, the model is a big part of that, but the harness is helping bring them into its own reach. So, 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, it's actually the harness which unlocks a new frontier for the model to then solve and kicks off that flywheel again. So, 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 pure coding. I'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. So, there's a lot of other things in the job that matter. Um this is the core idea that I would like you to understand, this flywheel and how the model and the harness together drive it. One 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 to solve the full software development life cycle. Now, 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. To 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 immediately auto triages that bug. Then 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. And 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. So, you can think of all this as almost like a meta harness, which is orchestrating many agents together. That's how I'd like you to think about this. So, 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. And then the next reason is there are some parts of the harness that the model can't absorb. So, let's talk about those. First of all, there is the infrastructure and runtime needed to make an autonomous colleague reliable. Over the past year plus, we've gone from agents that work on a typical task for minutes to hours to days and even longer. And there's just no way that this can work on one person's laptop, which might go to sleep. It has to be running in the cloud if you really want an autonomous colleague. And then next, we have trust and security. Yes, 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 to the model. Um or if you're okay with that, I'm sure that your head of security would have something to say. Um 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? And then finally, there's efficiency. If 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 that there was a significant difference in cost per task. I have personally been focused over the past month on optimizing the efficiency of the Grok Bot harness. Now, 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 for users without any hit to quality. Users have really, really been appreciating this. It's been awesome to see. Um 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. So, by now you might be convinced that the harness still matters. Then you might be asking, "What's next?" Here's a few trends that I'm anticipating. First 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. That means that the harness is going to get more messy before it gets cleaned up. There's going to be more things shipped. So, because of this, we're expecting people to um build their harnesses, design them so it's easier to ship and unship new features. One 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. Some of you may have heard of Hyrum's Law, which says that anything that you expose to users, they will come to depend on. Now, 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. Now, another way that we've been seeing um this trend manifest is a bigger focus on designing primitives for agents. Primitives are like building block interfaces that can be composed and extended into powerful features. When 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. It's just good engineering. One example across our agent harnesses of a primitive we've added is called subscriptions. Agents need a way to wake up in response to events. Those might be someone posting in Slack, CI going red on a PR that the agent cares about, or just a certain time of day. Now, 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. And then another cool thing that we do, I mentioned earlier that we have these deterministic guardrails to patch over bad behaviors in the model. That 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." Now, we'd rather not do these things. They'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. So, 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, and every time the model uses one, it gets logged. So, 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. Another thing that I think is next for the harness is it is moving up to a higher level of abstraction. Yesterday's generation of harnesses were built from composing a bunch of LLM calls. Now, the new harnesses we're seeing, like GrokBot and the software factory, are using agents as the core building block. And then finally, I predict that within 6 months, 90% of agent work is going to run off the laptop. This 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. So, is the harness dead? The parts that made it smarter at coding, yes, that is the flywheel at work. That is the flywheel at work. But, the harness, as we saw, is how the agents can unlock the next frontier of work. And then also the trust, the efficiency, the infrastructure that the harness brings, those things are not going anywhere. So, yesterday's harness may be dead, but long live the agent harness of tomorrow. Thank you. \>\> \[applause\]