Don't build more AI agents until you watch this

Vercel made its agent better by deleting Vercel made its agent better by deleting 80% of its tools. You heard that right. 80% of its tools. You heard that right. 80% of its tools. You heard that right. And that sentence can sound wrong if And that sentence can sound wrong if And that sentence can sound wrong if you've been following a lot of the hype you've been following a lot of the hype you've been following a lot of the hype around new tools and new skills for around new tools and new skills for around new tools and new skills for agents. So, I want to set the record agents. So, I want to set the record agents. So, I want to set the record straight. The usual story we hear is straight. The usual story we hear is straight. The usual story we hear is that agents get better as you give them that agents get better as you give them that agents get better as you give them more stuff, right? More context, more more stuff, right? More context, more more stuff, right? More context, more memory, more tools, more integrations, memory, more tools, more integrations, memory, more tools, more integrations, more access, more autonomy. Let the more access, more autonomy. Let the more access, more autonomy. Let the agent touch the CRM, let it use Slack, agent touch the CRM, let it use Slack, agent touch the CRM, let it use Slack, let it browse the web, let it update the let it browse the web, let it update the let it browse the web, let it update the record. Vercel's example is a really record. Vercel's example is a really record. Vercel's example is a really healthy counterexample in that process. healthy counterexample in that process. healthy counterexample in that process. And no, it's not just about context And no, it's not just about context And no, it's not just about context window, which is the usual reason people window, which is the usual reason people window, which is the usual reason people dump out tools. Messages came in. Some dump out tools. Messages came in. Some dump out tools. Messages came in. Some were real leads, some were spam, some were real leads, some were spam, some were real leads, some were spam, some were support questions dressed up as were support questions dressed up as were support questions dressed up as sales questions, some came from little sales questions, some came from little sales questions, some came from little companies, some came from accounts that companies, some came from accounts that companies, some came from accounts that might matter to Vercel, some deserved a might matter to Vercel, some deserved a might matter to Vercel, some deserved a really quick reply, and some needed really quick reply, and some needed really quick reply, and some needed research, and some needed routing. It's research, and some needed routing. It's research, and some needed routing. It's the usual messy inbox. So, Vercel the usual messy inbox. So, Vercel the usual messy inbox. So, Vercel studied one of its best reps. They studied one of its best reps. They studied one of its best reps. They watched the workflow closely enough to watched the workflow closely enough to watched the workflow closely enough to turn pieces of it into an agent. I love turn pieces of it into an agent. I love turn pieces of it into an agent. I love that part. You have to study what people that part. You have to study what people that part. You have to study what people are already doing. What did the rep are already doing. What did the rep are already doing. What did the rep ignore? What did they answer? What made ignore? What did they answer? What made ignore? What did they answer? What made a lead real? What research happened a lead real? What research happened a lead real? What research happened before the reply? When was a message before the reply? When was a message before the reply? When was a message actually a support issue? Where did the actually a support issue? Where did the actually a support issue? Where did the human still need to make a judgment human still need to make a judgment human still need to make a judgment call? Then, they built the agent around call? Then, they built the agent around call? Then, they built the agent around the actual observed workflow, not the the actual observed workflow, not the the actual observed workflow, not the paper workflow. Again, love that. The paper workflow. Again, love that. The paper workflow. Again, love that. The agent filtered inbound messages, it agent filtered inbound messages, it agent filtered inbound messages, it qualified leads, it researched qualified leads, it researched qualified leads, it researched companies, it drafted responses, it companies, it drafted responses, it companies, it drafted responses, it routed support questions away from routed support questions away from routed support questions away from sales. A human still reviewed the work sales. A human still reviewed the work sales. A human still reviewed the work because the goal was not to let a bot because the goal was not to let a bot because the goal was not to let a bot roam around the company, right? The goal roam around the company, right? The goal roam around the company, right? The goal was to take a repeatable workflow from a

was to take a repeatable workflow from a was to take a repeatable workflow from a strong employee and make that repeatable strong employee and make that repeatable strong employee and make that repeatable bit run fast. And that's already a great bit run fast. And that's already a great bit run fast. And that's already a great story. But, the more important lesson is story. But, the more important lesson is story. But, the more important lesson is what happened after the agent existed. what happened after the agent existed. what happened after the agent existed. The agent did not get better when the The agent did not get better when the The agent did not get better when the team kept piling on tools. It got better team kept piling on tools. It got better team kept piling on tools. It got better when they took away tools. And this is when they took away tools. And this is when they took away tools. And this is something that I think that a lot of something that I think that a lot of something that I think that a lot of folks who are excited about agents need folks who are excited about agents need folks who are excited about agents need to sit with more. And this goes for to sit with more. And this goes for to sit with more. And this goes for skills, too. If you've got a pile of skills, too. If you've got a pile of skills, too. If you've got a pile of skills in your codex or Claude, pay skills in your codex or Claude, pay skills in your codex or Claude, pay attention. Because most of us are attention. Because most of us are attention. Because most of us are building agents the opposite way in building agents the opposite way in building agents the opposite way in practice. We're so enthused about practice. We're so enthused about practice. We're so enthused about building, right? We start with one task building, right? We start with one task building, right? We start with one task and add a tool and add another tool and and add a tool and add another tool and and add a tool and add another tool and add a memory file and add a slack add a memory file and add a slack add a memory file and add a slack integration, add a browser, and add a integration, add a browser, and add a integration, add a browser, and add a CRM action, add another exception. And CRM action, add another exception. And CRM action, add another exception. And after a while, the agent will look super after a while, the agent will look super after a while, the agent will look super powerful and muscled up, but it's going powerful and muscled up, but it's going powerful and muscled up, but it's going to become harder to trust. The beginner to become harder to trust. The beginner to become harder to trust. The beginner instinct is to add. The maintenance instinct is to add. The maintenance instinct is to add. The maintenance instinct is to ask what should be instinct is to ask what should be instinct is to ask what should be removed. That is the real agent story of removed. That is the real agent story of removed. That is the real agent story of 2026. Not can you build an agent? Look, 2026. Not can you build an agent? Look, 2026. Not can you build an agent? Look, I've got a video on that. There's dozens I've got a video on that. There's dozens I've got a video on that. There's dozens of videos out there on that. Of course, of videos out there on that. Of course, of videos out there on that. Of course, you can build an agent. The harder you can build an agent. The harder you can build an agent. The harder question is whether you can keep the question is whether you can keep the question is whether you can keep the setup around the agent healthy as the setup around the agent healthy as the setup around the agent healthy as the work changes and the model evolves.

work changes and the model evolves. work changes and the model evolves. People call that setup a harness. If People call that setup a harness. If People call that setup a harness. If that word feels super technical, you can that word feels super technical, you can that word feels super technical, you can call it a workbench. It's kind of the call it a workbench. It's kind of the call it a workbench. It's kind of the same thing. The agent is the worker, the same thing. The agent is the worker, the same thing. The agent is the worker, the harness is the workbench. It's what the harness is the workbench. It's what the harness is the workbench. It's what the agent reads, it's what it remembers, agent reads, it's what it remembers, agent reads, it's what it remembers, it's what tools it can touch, it's what it's what tools it can touch, it's what it's what tools it can touch, it's what it's allowed to change, it's what proof it's allowed to change, it's what proof it's allowed to change, it's what proof it has to bring back, it's what stops it it has to bring back, it's what stops it it has to bring back, it's what stops it when the work gets risky. For Salesforce when the work gets risky. For Salesforce when the work gets risky. For Salesforce agent, had a workbench or harness. It agent, had a workbench or harness. It agent, had a workbench or harness. It had a documented workflow from a top had a documented workflow from a top had a documented workflow from a top performer. It had tools, it had performer. It had tools, it had performer. It had tools, it had handoffs, it had human review, it had handoffs, it had human review, it had handoffs, it had human review, it had feedback, and then the team learned that feedback, and then the team learned that feedback, and then the team learned that part of maintaining that workbench or part of maintaining that workbench or part of maintaining that workbench or harness is pruning. And that is a much harness is pruning. And that is a much harness is pruning. And that is a much more important lesson than AI replaced more important lesson than AI replaced more important lesson than AI replaced the sales process, which is what all the the sales process, which is what all the the sales process, which is what all the headlines were about. The real lesson is headlines were about. The real lesson is headlines were about. The real lesson is that useful agents desperately need good that useful agents desperately need good that useful agents desperately need good maintenance. And I think there are four maintenance. And I think there are four maintenance. And I think there are four first principles here that I want to lay first principles here that I want to lay first principles here that I want to lay out that are going to be durable for out that are going to be durable for out that are going to be durable for 2026. The first is that agents 2026. The first is that agents 2026. The first is that agents themselves are moving. The model themselves are moving. The model themselves are moving. The model underneath the agent is not stable. It's underneath the agent is not stable. It's underneath the agent is not stable. It's getting better, it's getting better at getting better, it's getting better at getting better, it's getting better at tool use, it's better at reasoning tool use, it's better at reasoning tool use, it's better at reasoning across steps, it's better at across steps, it's better at across steps, it's better at understanding messy instructions, it's understanding messy instructions, it's understanding messy instructions, it's better at reading files, it's better at better at reading files, it's better at better at reading files, it's better at remembering what matters, it's better at remembering what matters, it's better at remembering what matters, it's better at moving through work without needing moving through work without needing moving through work without needing every step spelled out. So, agents get every step spelled out. So, agents get every step spelled out. So, agents get better, and that sounds purely good. And better, and that sounds purely good. And better, and that sounds purely good. And mostly it is, but it also means mostly it is, but it also means mostly it is, but it also means yesterday's harness can become very yesterday's harness can become very yesterday's harness can become very wrong very quickly at the price of an wrong very quickly at the price of an wrong very quickly at the price of an update. A tool that helped a weaker update. A tool that helped a weaker update. A tool that helped a weaker model can confuse a stronger one. A A model can confuse a stronger one. A A model can confuse a stronger one. A A rule that protected you from an rule that protected you from an rule that protected you from an unreliable model's mistakes can trap a unreliable model's mistakes can trap a unreliable model's mistakes can trap a better model. A workflow that forced better model. A workflow that forced better model. A workflow that forced structure around a clumsy agent can

structure around a clumsy agent can structure around a clumsy agent can become a drag when the model can handle become a drag when the model can handle become a drag when the model can handle a lot more of the work itself. These are a lot more of the work itself. These are a lot more of the work itself. These are all real examples. We are used to all real examples. We are used to all real examples. We are used to software breaking when it gets worse. software breaking when it gets worse. software breaking when it gets worse. That's our mental model. That's our mental model. That's our mental model. Agents can also break when the model Agents can also break when the model Agents can also break when the model gets better and that is a different and gets better and that is a different and gets better and that is a different and new thing. It's a strange new new thing. It's a strange new new thing. It's a strange new maintenance problem. Imagine the first maintenance problem. Imagine the first maintenance problem. Imagine the first version of an agent is not very version of an agent is not very version of an agent is not very reliable. It overreaches. It invents reliable. It overreaches. It invents reliable. It overreaches. It invents patterns. It treats one example like a patterns. It treats one example like a patterns. It treats one example like a trend. So you build a really careful trend. So you build a really careful trend. So you build a really careful harness around it. You give it strict harness around it. You give it strict harness around it. You give it strict tools and narrow the prompt and say only tools and narrow the prompt and say only tools and narrow the prompt and say only use these sources. Don't infer. Don't use these sources. Don't infer. Don't use these sources. Don't infer. Don't create records. Don't recommend a next create records. Don't recommend a next create records. Don't recommend a next step. Just summarize what you see. And step. Just summarize what you see. And step. Just summarize what you see. And that may be exactly right for that that may be exactly right for that that may be exactly right for that model. model. model. the model improves. Again, real examples the model improves. Again, real examples the model improves. Again, real examples here. Now it can compare sources better. here. Now it can compare sources better. here. Now it can compare sources better. It can understand the workflow better. It can understand the workflow better. It can understand the workflow better. It can tell the difference between a It can tell the difference between a It can tell the difference between a weak signal and a real pattern. It can weak signal and a real pattern. It can weak signal and a real pattern. It can draft a useful next step. I am draft a useful next step. I am draft a useful next step. I am describing November to March of this describing November to March of this describing November to March of this past 6 or 8 months. But your harness past 6 or 8 months. But your harness past 6 or 8 months. But your harness still treats it like the old model. So still treats it like the old model. So still treats it like the old model. So the agent is underused or the opposite the agent is underused or the opposite the agent is underused or the opposite happens, right? The old model was clumsy happens, right? The old model was clumsy happens, right? The old model was clumsy so you give it broad access because you so you give it broad access because you so you give it broad access because you knew a human would catch everything.

knew a human would catch everything. knew a human would catch everything. Then the model gets better. Now it can Then the model gets better. Now it can Then the model gets better. Now it can take 20 plausible actions in a few take 20 plausible actions in a few take 20 plausible actions in a few minutes. Now they look real. They look minutes. Now they look real. They look minutes. Now they look real. They look organized. They create work that a human organized. They create work that a human organized. They create work that a human has to unwind. So the model improved, has to unwind. So the model improved, has to unwind. So the model improved, the harness did not and that is a the harness did not and that is a the harness did not and that is a massive driver of agent breakage in massive driver of agent breakage in massive driver of agent breakage in 2026. Normal systems drift. Prompts 2026. Normal systems drift. Prompts 2026. Normal systems drift. Prompts drift. Wiki's get stale. Dashboards drift. Wiki's get stale. Dashboards drift. Wiki's get stale. Dashboards break. Automations keep running long break. Automations keep running long break. Automations keep running long after the process changes. SOPs describe after the process changes. SOPs describe after the process changes. SOPs describe how the company worked months ago. Slack how the company worked months ago. Slack how the company worked months ago. Slack channels become junk drawers. Templates channels become junk drawers. Templates channels become junk drawers. Templates survive long after the reason for the survive long after the reason for the survive long after the reason for the template disappeared. None of that template disappeared. None of that template disappeared. None of that started with AI, right? Every company started with AI, right? Every company started with AI, right? Every company already has this problem. The product already has this problem. The product already has this problem. The product wiki, it's a little or a lot wrong. The wiki, it's a little or a lot wrong. The wiki, it's a little or a lot wrong. The CRM field means something slightly CRM field means something slightly CRM field means something slightly different than it used to. The different than it used to. The different than it used to. The dashboard, it still says activation, but dashboard, it still says activation, but dashboard, it still says activation, but the team changed what activation means. the team changed what activation means. the team changed what activation means. The support tags have evolved. The road The support tags have evolved. The road The support tags have evolved. The road map moved, the owner changed, the map moved, the owner changed, the map moved, the owner changed, the process changed, the docs didn't. With process changed, the docs didn't. With process changed, the docs didn't. With normal software, this is vaguely normal software, this is vaguely normal software, this is vaguely annoying and you sometimes get messages annoying and you sometimes get messages annoying and you sometimes get messages saying, "Please update your wiki." With saying, "Please update your wiki." With saying, "Please update your wiki." With agents, it's very dangerous because agents, it's very dangerous because agents, it's very dangerous because agents don't sit. They produce work. agents don't sit. They produce work. agents don't sit. They produce work. They're proactive. That's their job.

They're proactive. That's their job. They're proactive. That's their job. They summarize, they recommend, they They summarize, they recommend, they They summarize, they recommend, they draft, they route, they update, and draft, they route, they update, and draft, they route, they update, and sometimes, of course, they act. That's sometimes, of course, they act. That's sometimes, of course, they act. That's the value. So, a stale wiki that is the value. So, a stale wiki that is the value. So, a stale wiki that is annoying to you is incredibly dangerous annoying to you is incredibly dangerous annoying to you is incredibly dangerous to an agent because it doesn't know that to an agent because it doesn't know that to an agent because it doesn't know that and it just keeps on working. And this and it just keeps on working. And this and it just keeps on working. And this is the second principle I want to is the second principle I want to is the second principle I want to communicate. Agents inherit all of the communicate. Agents inherit all of the communicate. Agents inherit all of the crud of the systems around them. If your crud of the systems around them. If your crud of the systems around them. If your wiki is stale, your agent reads and wiki is stale, your agent reads and wiki is stale, your agent reads and ingests stale truth. If your process ingests stale truth. If your process ingests stale truth. If your process changed, your agent will follow old changed, your agent will follow old changed, your agent will follow old process unless you update your docs. If process unless you update your docs. If process unless you update your docs. If your prompt is written for last your prompt is written for last your prompt is written for last quarter's company and model, that agent quarter's company and model, that agent quarter's company and model, that agent may keep serving last quarter's company may keep serving last quarter's company may keep serving last quarter's company and not realize everything's changed. and not realize everything's changed. and not realize everything's changed. If your dashboard definition is If your dashboard definition is If your dashboard definition is incorrect now, the agent will make the incorrect now, the agent will make the incorrect now, the agent will make the wrong number feel very convincing. This wrong number feel very convincing. This wrong number feel very convincing. This is not a model failure in the simple is not a model failure in the simple is not a model failure in the simple sense, right? The agent did its job. sense, right? The agent did its job. sense, right? The agent did its job. It's the old maintenance problem with a It's the old maintenance problem with a It's the old maintenance problem with a machine that now can produce work from machine that now can produce work from machine that now can produce work from that mess that is sometimes very that mess that is sometimes very that mess that is sometimes very convincing. And this is why Stewart convincing. And this is why Stewart convincing. And this is why Stewart Brand's Maintenance of Everything, I Brand's Maintenance of Everything, I Brand's Maintenance of Everything, I think it's the right frame for agents.

think it's the right frame for agents. think it's the right frame for agents. Brand is writing about sailboats and Brand is writing about sailboats and Brand is writing about sailboats and vehicles and weapons and manuals and vehicles and weapons and manuals and vehicles and weapons and manuals and corrosion and the work that keeps corrosion and the work that keeps corrosion and the work that keeps important systems alive after the launch important systems alive after the launch important systems alive after the launch moment is over. Agents are a lot less moment is over. Agents are a lot less moment is over. Agents are a lot less like apps and more like sailboats. I like apps and more like sailboats. I like apps and more like sailboats. I love this book. This is like one of my love this book. This is like one of my love this book. This is like one of my favorite books of the year. You don't favorite books of the year. You don't favorite books of the year. You don't just launch agents and walk away. The just launch agents and walk away. The just launch agents and walk away. The weather changes, the lines loosen, salt weather changes, the lines loosen, salt weather changes, the lines loosen, salt gets into everything, and yes, this is gets into everything, and yes, this is gets into everything, and yes, this is all from that book. The same setup that all from that book. The same setup that all from that book. The same setup that worked yesterday can be wrong tomorrow. worked yesterday can be wrong tomorrow. worked yesterday can be wrong tomorrow. A sailboat is not maintained because it A sailboat is not maintained because it A sailboat is not maintained because it was badly designed, it is maintained was badly designed, it is maintained was badly designed, it is maintained because it lives in motion. Agents live because it lives in motion. Agents live because it lives in motion. Agents live in motion, too. in motion, too. in motion, too. The model changes inside them. The world The model changes inside them. The world The model changes inside them. The world changes around them. In that sense, they changes around them. In that sense, they changes around them. In that sense, they are much more like traditional vehicle are much more like traditional vehicle are much more like traditional vehicle maintenance than anything else we've maintenance than anything else we've maintenance than anything else we've seen in software in a long time. The seen in software in a long time. The seen in software in a long time. The harness has to keep up with the model harness has to keep up with the model harness has to keep up with the model changes and the world changes. changes and the world changes. changes and the world changes. And so few of us really have a good And so few of us really have a good And so few of us really have a good system for that. Now, the third system for that. Now, the third system for that. Now, the third principle I want to call out is that the principle I want to call out is that the principle I want to call out is that the biggest AI companies already know this. biggest AI companies already know this. biggest AI companies already know this. A lot of the implicit bet from the A lot of the implicit bet from the A lot of the implicit bet from the frontier labs and platform companies is frontier labs and platform companies is frontier labs and platform companies is not just that their models will get not just that their models will get not just that their models will get better. It is that they can use those better. It is that they can use those better. It is that they can use those better models to ship and evolve the better models to ship and evolve the better models to ship and evolve the harness faster. And I think that's one harness faster. And I think that's one harness faster. And I think that's one reason why it's really important to talk reason why it's really important to talk reason why it's really important to talk about Codex in the strategic context of about Codex in the strategic context of about Codex in the strategic context of OpenAI's long-term strategy. And I think OpenAI's long-term strategy. And I think OpenAI's long-term strategy. And I think that's one reason Codex matters so much.

that's one reason Codex matters so much. that's one reason Codex matters so much. Codex is strong not just because the Codex is strong not just because the Codex is strong not just because the model is strong. Codex is strong because model is strong. Codex is strong because model is strong. Codex is strong because OpenAI keeps maintaining the harness OpenAI keeps maintaining the harness OpenAI keeps maintaining the harness around the model so it feels intuitive around the model so it feels intuitive around the model so it feels intuitive and native as the model and the world and native as the model and the world and native as the model and the world evolve around it. It has become closer evolve around it. It has become closer evolve around it. It has become closer to an operating surface for work as it's to an operating surface for work as it's to an operating surface for work as it's evolved. So, it has a terminal and a evolved. So, it has a terminal and a evolved. So, it has a terminal and a desktop app and an IDE and a browser and desktop app and an IDE and a browser and desktop app and an IDE and a browser and computer use and files and plugins and computer use and files and plugins and computer use and files and plugins and memory and automations and approvals and memory and automations and approvals and memory and automations and approvals and sandboxing and network controls and sandboxing and network controls and sandboxing and network controls and keychain storage and manage configs and keychain storage and manage configs and keychain storage and manage configs and logs. logs. logs. This is way beyond a chat box with a This is way beyond a chat box with a This is way beyond a chat box with a smarter brain. It's a very carefully smarter brain. It's a very carefully smarter brain. It's a very carefully maintained workbench around machine maintained workbench around machine maintained workbench around machine work. And the Claude code team is doing work. And the Claude code team is doing work. And the Claude code team is doing the same thing, right? They're investing the same thing, right? They're investing the same thing, right? They're investing heavily in their harness. I'm really heavily in their harness. I'm really heavily in their harness. I'm really excited to do the Claude code is as excited to do the Claude code is as excited to do the Claude code is as amazing as Codex review, guys. So, amazing as Codex review, guys. So, amazing as Codex review, guys. So, please give me something that cool. And please give me something that cool. And please give me something that cool. And to go back to the workbench analogy, to go back to the workbench analogy, to go back to the workbench analogy, every tool in that workbench is every tool in that workbench is every tool in that workbench is carefully chosen with Codex, right? The carefully chosen with Codex, right? The carefully chosen with Codex, right? The terminal matters because real work lives terminal matters because real work lives terminal matters because real work lives in commands and repos and files and in commands and repos and files and in commands and repos and files and tests and local tools. This was Clyde tests and local tools. This was Clyde tests and local tools. This was Clyde Code's original insight, by the way. The Code's original insight, by the way. The Code's original insight, by the way. The browser matters because real work browser matters because real work browser matters because real work happens on interfaces that humans see.

happens on interfaces that humans see. happens on interfaces that humans see. And both Anthropic and OpenAI are And both Anthropic and OpenAI are And both Anthropic and OpenAI are building that way. Computer use matters building that way. Computer use matters building that way. Computer use matters because not every tool has a clean API. because not every tool has a clean API. because not every tool has a clean API. Plugins matter because work lives in a Plugins matter because work lives in a Plugins matter because work lives in a bunch of other systems: GitHub, Google bunch of other systems: GitHub, Google bunch of other systems: GitHub, Google Drive, Jira, Slack, etc. Memory matters Drive, Jira, Slack, etc. Memory matters Drive, Jira, Slack, etc. Memory matters because preferences and corrections because preferences and corrections because preferences and corrections should not have to be rebuilt every day, should not have to be rebuilt every day, should not have to be rebuilt every day, right? Approvals and sandboxing matter right? Approvals and sandboxing matter right? Approvals and sandboxing matter because a capable agent still needs because a capable agent still needs because a capable agent still needs boundaries. Logs matter because when an boundaries. Logs matter because when an boundaries. Logs matter because when an agent does something weird, someone agent does something weird, someone agent does something weird, someone needs to know what happened. This whole needs to know what happened. This whole needs to know what happened. This whole surface together is the harness. It's an surface together is the harness. It's an surface together is the harness. It's an art to build a good harness. And there art to build a good harness. And there art to build a good harness. And there are really two teams in the world are really two teams in the world are really two teams in the world building good harnesses: the Anthropic building good harnesses: the Anthropic building good harnesses: the Anthropic team and the OpenAI team right now. And team and the OpenAI team right now. And team and the OpenAI team right now. And this is where the hyperscaler and this is where the hyperscaler and this is where the hyperscaler and frontier platform bet gets super frontier platform bet gets super frontier platform bet gets super interesting. If the model can help you interesting. If the model can help you interesting. If the model can help you ship the harness and test the harness ship the harness and test the harness ship the harness and test the harness and refactor the harness and observe the and refactor the harness and observe the and refactor the harness and observe the harness and train the harness, then harness and train the harness, then harness and train the harness, then capability gain is going to start to capability gain is going to start to capability gain is going to start to compound real fast. Because better compound real fast. Because better compound real fast. Because better agents can help build more effective agents can help build more effective agents can help build more effective harnesses, better harnesses can make the harnesses, better harnesses can make the harnesses, better harnesses can make the agents more useful, and then better agents more useful, and then better agents more useful, and then better agents can help rebuild that harness agents can help rebuild that harness agents can help rebuild that harness once more. That is why the Vercel story once more. That is why the Vercel story once more. That is why the Vercel story is not just a quirky sales automation is not just a quirky sales automation is not just a quirky sales automation story. It's a pattern we all need to story. It's a pattern we all need to story. It's a pattern we all need to learn from. The companies that win are learn from. The companies that win are learn from. The companies that win are not the ones that build the perfect not the ones that build the perfect not the ones that build the perfect wrapper once. They're the ones that keep wrapper once. They're the ones that keep wrapper once. They're the ones that keep rebuilding the wrapper as the model and rebuilding the wrapper as the model and rebuilding the wrapper as the model and the work change. They rebuild that the work change. They rebuild that the work change. They rebuild that workshop. They rebuild that harness. And workshop. They rebuild that harness. And workshop. They rebuild that harness. And this is why the direction of Codex over this is why the direction of Codex over this is why the direction of Codex over time feels really significant to me. If time feels really significant to me. If time feels really significant to me. If Codex keeps getting more capable, and Codex keeps getting more capable, and Codex keeps getting more capable, and that's an if, that's an if, that's an if, and the Codex harness keeps getting and the Codex harness keeps getting and the Codex harness keeps getting closer to the operating system of work, closer to the operating system of work, closer to the operating system of work, another if, then OpenAI is not only

another if, then OpenAI is not only another if, then OpenAI is not only selling intelligence, they are selling selling intelligence, they are selling selling intelligence, they are selling the environment in which intelligence the environment in which intelligence the environment in which intelligence becomes useful. And that's the same bet becomes useful. And that's the same bet becomes useful. And that's the same bet Anthropic is making with Clyde Code and Anthropic is making with Clyde Code and Anthropic is making with Clyde Code and Clyde Co-work. The harness evolves, the Clyde Co-work. The harness evolves, the Clyde Co-work. The harness evolves, the model evolves, the harness lets the model evolves, the harness lets the model evolves, the harness lets the model touch more real work over time. model touch more real work over time. model touch more real work over time. More real work creates more pressure to More real work creates more pressure to More real work creates more pressure to improve the harness, and that loop is improve the harness, and that loop is improve the harness, and that loop is ignited like a flywheel. And that loop ignited like a flywheel. And that loop ignited like a flywheel. And that loop matters, and it raises the bar for all matters, and it raises the bar for all matters, and it raises the bar for all of the rest of us. Because if you're of the rest of us. Because if you're of the rest of us. Because if you're building your own agent setup, you are building your own agent setup, you are building your own agent setup, you are now not just choosing a model, you're now not just choosing a model, you're now not just choosing a model, you're choosing how much harness maintenance choosing how much harness maintenance choosing how much harness maintenance you are choosing to own versus how much you are choosing to own versus how much you are choosing to own versus how much harness maintenance you're outsourcing. harness maintenance you're outsourcing. harness maintenance you're outsourcing. A light custom harness might be a clean A light custom harness might be a clean A light custom harness might be a clean set of instructions and memory and set of instructions and memory and set of instructions and memory and source folders and repeatable methods source folders and repeatable methods source folders and repeatable methods around Codex or Quad. That can be around Codex or Quad. That can be around Codex or Quad. That can be enough. enough. enough. Here are the sources. Here's the job. Here are the sources. Here's the job. Here are the sources. Here's the job. Here's what you can't touch. Here's the Here's what you can't touch. Here's the Here's what you can't touch. Here's the proof I need. Here's when a human proof I need. Here's when a human proof I need. Here's when a human decides. A deeper custom harness is a decides. A deeper custom harness is a decides. A deeper custom harness is a very different thing. Because now you very different thing. Because now you very different thing. Because now you have a data feed, a review screen, have a data feed, a review screen, have a data feed, a review screen, permission levels, logs, model choice, permission levels, logs, model choice, permission levels, logs, model choice, escalation paths, approval rules, and a escalation paths, approval rules, and a escalation paths, approval rules, and a plan for what happens when the model plan for what happens when the model plan for what happens when the model changes. And that can be very worth it changes. And that can be very worth it changes. And that can be very worth it to invest in, but now you're not just to invest in, but now you're not just to invest in, but now you're not just building an agent. You are investing in building an agent. You are investing in building an agent. You are investing in the long-term maintenance of an agent the long-term maintenance of an agent the long-term maintenance of an agent and harness system. You are taking and harness system. You are taking and harness system. You are taking responsibility for evolving the system responsibility for evolving the system responsibility for evolving the system around the agent over time. And the more around the agent over time. And the more around the agent over time. And the more custom the harness, the more you own the custom the harness, the more you own the custom the harness, the more you own the upkeep. And this is not abstract for me.

upkeep. And this is not abstract for me. upkeep. And this is not abstract for me. So now I'm thinking about my delegation So now I'm thinking about my delegation So now I'm thinking about my delegation model differently, and part of it is model differently, and part of it is model differently, and part of it is just the ordinary mess of work, right? just the ordinary mess of work, right? just the ordinary mess of work, right? Folders move, drafts change, source Folders move, drafts change, source Folders move, drafts change, source packets get updated, memory gets stale, packets get updated, memory gets stale, packets get updated, memory gets stale, and the way I want the agent to use and the way I want the agent to use and the way I want the agent to use local context changes as the agent gets local context changes as the agent gets local context changes as the agent gets better. So the thing I maintain is a lot better. So the thing I maintain is a lot better. So the thing I maintain is a lot more than a prompt. It's It's the whole more than a prompt. It's It's the whole more than a prompt. It's It's the whole way the agent meets my files. way the agent meets my files. way the agent meets my files. Where should it look first? Which Where should it look first? Which Where should it look first? Which folders are a source of truth? What folders are a source of truth? What folders are a source of truth? What should it ignore? What should it ask should it ignore? What should it ask should it ignore? What should it ask about before touching? What should it about before touching? What should it about before touching? What should it remember? What should it forget? When it remember? What should it forget? When it remember? What should it forget? When it searches memory, searches memory, searches memory, is that right? When does it actually go is that right? When does it actually go is that right? When does it actually go read the file? That is a harness read the file? That is a harness read the file? That is a harness question for me. And that's a tiny question for me. And that's a tiny question for me. And that's a tiny personal harness question, right? I'm personal harness question, right? I'm personal harness question, right? I'm not even talking about team harnesses not even talking about team harnesses not even talking about team harnesses here. And it has changed because the here. And it has changed because the here. And it has changed because the agents have changed, because the models agents have changed, because the models agents have changed, because the models have updated. And have updated. And have updated. And this brings me this brings me this brings me to the fourth principle, and it's the to the fourth principle, and it's the to the fourth principle, and it's the one that I think matters the most. You one that I think matters the most. You one that I think matters the most. You need to ask, I think all of us need to need to ask, I think all of us need to need to ask, I think all of us need to ask, what is my harness? What is my ask, what is my harness? What is my ask, what is my harness? What is my workshop? Not in a sort of technical way workshop? Not in a sort of technical way workshop? Not in a sort of technical way that makes it feel scary, but in a very that makes it feel scary, but in a very that makes it feel scary, but in a very practical way. If you use chat GPT or practical way. If you use chat GPT or practical way. If you use chat GPT or Claude or or CodeX or any other agentic Claude or or CodeX or any other agentic Claude or or CodeX or any other agentic tool, your harness is the setup that tool, your harness is the setup that tool, your harness is the setup that makes that model useful for your real makes that model useful for your real makes that model useful for your real work. Maybe it's project folders, maybe work. Maybe it's project folders, maybe work. Maybe it's project folders, maybe it's your memory, your prompts, your it's your memory, your prompts, your it's your memory, your prompts, your source docs, your approval habits, your source docs, your approval habits, your source docs, your approval habits, your browser access, your file rules, your browser access, your file rules, your browser access, your file rules, your tools, your verification loop, your way tools, your verification loop, your way tools, your verification loop, your way of asking for proof, your habit of of asking for proof, your habit of of asking for proof, your habit of making the agent read the actual source making the agent read the actual source making the agent read the actual source instead of guessing from context. If instead of guessing from context. If instead of guessing from context. If you're a product leader, your harness you're a product leader, your harness you're a product leader, your harness might be the sources your agent reads might be the sources your agent reads might be the sources your agent reads before planning, right? If you're in before planning, right? If you're in before planning, right? If you're in sales, your harness might be the CRM

sales, your harness might be the CRM sales, your harness might be the CRM fields or the call notes or the routing fields or the call notes or the routing fields or the call notes or the routing rules and and the human approval steps. rules and and the human approval steps. rules and and the human approval steps. Maybe you're in support, right? Your Maybe you're in support, right? Your Maybe you're in support, right? Your harness might be the policy store and harness might be the policy store and harness might be the policy store and escalation paths and refund rules. If escalation paths and refund rules. If escalation paths and refund rules. If you're a writer, your harness might be you're a writer, your harness might be you're a writer, your harness might be your drafts and your transcripts and your drafts and your transcripts and your drafts and your transcripts and your voice notes and your editorial your voice notes and your editorial your voice notes and your editorial rules and your archive and the rules and your archive and the rules and your archive and the instruction that the agent has to show instruction that the agent has to show instruction that the agent has to show where an idea came from. where an idea came from. where an idea came from. If you're an engineer, of course, your If you're an engineer, of course, your If you're an engineer, of course, your harness would be a repo, test, terminal, harness would be a repo, test, terminal, harness would be a repo, test, terminal, permissions, work trees, logs, review permissions, work trees, logs, review permissions, work trees, logs, review rules. That in many ways is the most rules. That in many ways is the most rules. That in many ways is the most mature example of a harness we have mature example of a harness we have mature example of a harness we have today. But that's the real question for today. But that's the real question for today. But that's the real question for all of us, whether we're engineers or all of us, whether we're engineers or all of us, whether we're engineers or not. What is your harness? What are you not. What is your harness? What are you not. What is your harness? What are you doing to ship it? What are you doing to doing to ship it? What are you doing to doing to ship it? What are you doing to rebuild it? Do you know when to rebuild rebuild it? Do you know when to rebuild rebuild it? Do you know when to rebuild it? Do you know how to rebuild it? What it? Do you know how to rebuild it? What it? Do you know how to rebuild it? What are you doing to evolve it as the models are you doing to evolve it as the models are you doing to evolve it as the models improve? The more useful question is improve? The more useful question is improve? The more useful question is can I maintain this sailboat over a can I maintain this sailboat over a can I maintain this sailboat over a voyage? What harness does this agent voyage? What harness does this agent voyage? What harness does this agent need? Those are the same question. And need? Those are the same question. And need? Those are the same question. And then the really mature question is then the really mature question is then the really mature question is what part of this harness will I need to what part of this harness will I need to what part of this harness will I need to delete later? And and Brand gets into delete later? And and Brand gets into delete later? And and Brand gets into this, he talks about simplicity as a key this, he talks about simplicity as a key this, he talks about simplicity as a key to maintenance, and I love that. This is to maintenance, and I love that. This is to maintenance, and I love that. This is the Vercel lesson. The agent got better the Vercel lesson. The agent got better the Vercel lesson. The agent got better when the workbench got cleaned up. And when the workbench got cleaned up. And when the workbench got cleaned up. And once you start to realize that your your once you start to realize that your your once you start to realize that your your maintenance question is not just a maintenance question is not just a maintenance question is not just a modification question, but potentially a modification question, but potentially a modification question, but potentially a deletion question, the whole agent deletion question, the whole agent deletion question, the whole agent conversation becomes a lot more complex.

conversation becomes a lot more complex. conversation becomes a lot more complex. You stop treating the harness as a You stop treating the harness as a You stop treating the harness as a one-time wrapper, you treat it as a one-time wrapper, you treat it as a one-time wrapper, you treat it as a living system where you do have to add living system where you do have to add living system where you do have to add sometimes, and you do have to take away. sometimes, and you do have to take away. sometimes, and you do have to take away. You have to think about the system's You have to think about the system's You have to think about the system's health overall. So, for any serious health overall. So, for any serious health overall. So, for any serious agent, I would check these five things. agent, I would check these five things. agent, I would check these five things. First, what's it eating, right? What's First, what's it eating, right? What's First, what's it eating, right? What's it reading? Are the sources current? Did it reading? Are the sources current? Did it reading? Are the sources current? Did the workflow move? Did a new source the workflow move? Did a new source the workflow move? Did a new source become important? Did an old source become important? Did an old source become important? Did an old source become misleading? become misleading? become misleading? Second, I would test its reach. What can Second, I would test its reach. What can Second, I would test its reach. What can it touch? Can it only read? Can it it touch? Can it only read? Can it it touch? Can it only read? Can it draft? Can it create tickets? Can it draft? Can it create tickets? Can it draft? Can it create tickets? Can it post in Slack? Can it update records? post in Slack? Can it update records? post in Slack? Can it update records? Can it spend money? Can it publish? A Can it spend money? Can it publish? A Can it spend money? Can it publish? A permission that was harmless for a permission that was harmless for a permission that was harmless for a weaker model may be too broad for a weaker model may be too broad for a weaker model may be too broad for a strong one. A restriction that made strong one. A restriction that made strong one. A restriction that made sense for an unreliable model may hold sense for an unreliable model may hold sense for an unreliable model may hold back a better one. Third, I would check back a better one. Third, I would check back a better one. Third, I would check its job. Is this still a summary agent? its job. Is this still a summary agent? its job. Is this still a summary agent? Is that useful? Is it becoming a Is that useful? Is it becoming a Is that useful? Is it becoming a planning agent because agents are planning agent because agents are planning agent because agents are getting better and it can be? Is it getting better and it can be? Is it getting better and it can be? Is it supposed to find themes? Is it supposed supposed to find themes? Is it supposed supposed to find themes? Is it supposed to recommend tradeoffs? Is it supposed to recommend tradeoffs? Is it supposed to recommend tradeoffs? Is it supposed to route work? Do not let the job change to route work? Do not let the job change to route work? Do not let the job change silently. Change the job on purpose if silently. Change the job on purpose if silently. Change the job on purpose if you're going to do it at all. Fourth, you're going to do it at all. Fourth, you're going to do it at all. Fourth, check the proof. The agent shouldn't check the proof. The agent shouldn't check the proof. The agent shouldn't just say, "Customers are frustrated with just say, "Customers are frustrated with just say, "Customers are frustrated with onboarding." It needs to link to the onboarding." It needs to link to the onboarding." It needs to link to the tickets. It should link to sales notes.

tickets. It should link to sales notes. tickets. It should link to sales notes. It should quote customer language and It should quote customer language and It should quote customer language and have a source. have a source. have a source. And it should say which sources it And it should say which sources it And it should say which sources it checked and where, and which ones it checked and where, and which ones it checked and where, and which ones it could not access. So, the proof is not could not access. So, the proof is not could not access. So, the proof is not just the agent saying it. The proof is just the agent saying it. The proof is just the agent saying it. The proof is an a linkable trail a human can inspect. an a linkable trail a human can inspect. an a linkable trail a human can inspect. Fifth and last, check the agent's value. Fifth and last, check the agent's value. Fifth and last, check the agent's value. I don't think this gets done enough. I don't think this gets done enough. I don't think this gets done enough. It's like asking if the sailboat got It's like asking if the sailboat got It's like asking if the sailboat got there, right? Does anyone read the there, right? Does anyone read the there, right? Does anyone read the output? Does it change the work? Does it output? Does it change the work? Does it output? Does it change the work? Does it save time after review? Does it create save time after review? Does it create save time after review? Does it create another pile of work? Is it duplicating another pile of work? Is it duplicating another pile of work? Is it duplicating a report? Has the model improved enough a report? Has the model improved enough a report? Has the model improved enough that the agent ought to be rebuilt? Has that the agent ought to be rebuilt? Has that the agent ought to be rebuilt? Has the business changed enough that the the business changed enough that the the business changed enough that the agent should be retired? That can agent should be retired? That can agent should be retired? That can happen. Agents, unlike almost anything happen. Agents, unlike almost anything happen. Agents, unlike almost anything else, break in two directions. They else, break in two directions. They else, break in two directions. They break because the world around them break because the world around them break because the world around them drifts, and they break because the model drifts, and they break because the model drifts, and they break because the model inside them improves. And maintenance is inside them improves. And maintenance is inside them improves. And maintenance is the work of keeping the harness fit the work of keeping the harness fit the work of keeping the harness fit between those two moving things. That between those two moving things. That between those two moving things. That delicate art, that sailing art, is the delicate art, that sailing art, is the delicate art, that sailing art, is the future of agents. It's not just more future of agents. It's not just more future of agents. It's not just more capability, it's better maintained capability, it's better maintained capability, it's better maintained capability. The work changes around the capability. The work changes around the capability. The work changes around the agent, the model changes inside it. And agent, the model changes inside it. And agent, the model changes inside it. And if you ignore either of those, the agent if you ignore either of those, the agent if you ignore either of those, the agent does not have to fail very loudly to does not have to fail very loudly to does not have to fail very loudly to become quite dangerous. All the agent become quite dangerous. All the agent become quite dangerous. All the agent has to do is to keep working, and it has to do is to keep working, and it has to do is to keep working, and it will start to haunt your business. And will start to haunt your business. And will start to haunt your business. And before I forget, yes, I'm going to say before I forget, yes, I'm going to say before I forget, yes, I'm going to say it again, read the Maintenance of it again, read the Maintenance of it again, read the Maintenance of Everything. If I could recommend one Everything. If I could recommend one Everything. If I could recommend one book on agents that isn't about AI, I book on agents that isn't about AI, I book on agents that isn't about AI, I would recommend this one. It's a would recommend this one. It's a would recommend this one. It's a phenomenal book. It's out of Stripe phenomenal book. It's out of Stripe phenomenal book. It's out of Stripe Press. I love what Stripe Press is Press. I love what Stripe Press is Press. I love what Stripe Press is doing. Thank you, guys. It's by Stewart doing. Thank you, guys. It's by Stewart doing. Thank you, guys. It's by Stewart Brand. Go get it and read it. It will Brand. Go get it and read it. It will Brand. Go get it and read it. It will teach you a lot about how to think about

teach you a lot about how to think about teach you a lot about how to think about the maintenance of technical systems. the maintenance of technical systems. the maintenance of technical systems. Have fun. I'll see you next time.