世界で最もAIに染まった会社の働き方：Spawnのエージェント艦隊

Imagine that your company was so AI native that agents were doing basically everything other than setting long-term vision and direction. We're talking spending 10 times the amount on compute and AI that you are on salaries and agentic fleets that are able to push 77,000 changes a month. That is the actual story of one company who has recently shared their playbook for how to become truly AI native. On today's episode, we are digging into those two years of hard-won knowledge and asking how it can apply to lots of other types of companies as well. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, C-Data, Granola, and Blitzy. To get an ad-free version of the show, go to patreon.com slash AI Daily Brief, or you can subscribe on Apple Podcasts. To learn more about sponsoring the show, send us a note at sponsors at aidailybrief.ai. And finally today, if you haven't had a chance to yet, check out my new podcast, How We Got

Here. It is a one hour deep dive that goes way farther and deeper in exploring modern trends than I have room to on this show. The first episode was about how tech lost America, the 15 year decline in trust that now sets up how the AI industry interfaces with society. And in a couple of days, I'll put out episode two, which is about how we made the wrong things cheap. You can find that on any of your podcast apps or YouTube or go to howwegothere.com. Welcome back to the AI Daily Brief Headlines Edition. all the daily AI news you need in around five minutes. We've got a really interesting one to kick us off. An AI startup is trying to buy a fallen white shoe law firm out of bankruptcy. Earlier this week, 95-year-old law firm Wild, Gottschall, and Manjiz collapsed and filed for Chapter 11 bankruptcy protection. They were one of the more storied law firms on Wall Street, among the top 20 firms in the country, and famously handling the bankruptcies of Lehman brothers and Enron. And while it's tempting to read an AI disruption story into their collapse, that just doesn't seem to be the case. The New York Times write-up instead blamed, quote,

an exodus of high-profile lawyers, a failed succession plan, and a scuttled deal to merge with a rival firm. In other words, yes, there has been pressure on big law to adopt AI and pass the savings onto their large corporate customers, but that is not the core story in this particular case. Which doesn't mean that it doesn't have implications and hints to show us where things are going for Legal AI. Because the information reports that legal AI startup Crosby has submitted a letter of intent to purchase some of Wiles' assets out of bankruptcy, including the trademark and brand as well as their internal data. Sources said that Crosby also hopes to hire some of Wiles' 1,200 lawyers. The reporting says it's unclear whether any other bidder has emerged for the fallen legal giant, which makes a lot of sense for anyone familiar with big law. Conventional wisdom is that law firms have approximately zero equity value of their own. all of the value is concentrated in the partners, frequently colloquially known as rainmakers who bring in all of the fees. Once those rainmakers walk, which is what happened with Weil, a law firm is just an office building with a bunch of very expensive support staff. That's what makes this bid from Crosby so interesting.

Over the past year, there's been a booming business for data brokers, who sell a variety of training data to AI startups for training agents, while his corporate records, which would have historically been valued at zero, could be worth tens of millions of dollars to the right buyer. Meanwhile, Crosby itself is still relatively small, last valued at $400 million and employing around 50 lawyers. Unlike other legal AI startups, Crosby operates as an AI-enhanced law firm with their own clients rather than selling their tools to other firms. That said, they're backed by Sequoia, Bain Capital, Lux Capital, and Index Ventures, so could have deep enough pockets or at least enough access to capital to make this deal work. If the play works out, Crosby could have the top quality data and brand required to make a run at building the first agentic white shoe law firm. And just to be super clear about what they'd be buying here, this isn't about while selling private client records, as that would be highly illegal. But law firms have an archive of precedent documents, contract templates, and workflows that provide a full playbook for completing legal work. In the case of while, this is a highly specialized category of legal work in bankruptcy law for

large corporates. The work is extremely expensive, as only a small handful of top firms have the institutional knowledge to complete it. Harneet Kay, a lawyer working on AI tools for distressed credit and restructuring, clocked this opportunity early, commenting on Wednesday, If while truly enters wind down, its data would be a seminal buy for an AI for law workflow startup, Harvey, Lagora, etc., and exceptionally valuable for restructuring and distressed workflows. There is a lot of data at 100-year-old firms like these, institutional knowledge, workflows, precedents, and proprietary data, that don't appear valuable on the surface, but could be extremely valuable to an AI startup in RL training. She noted that Google recently bid $10 million for the operational data from Spirit Airlines, which wouldn't be anywhere near as specialized as Wiles data. After news broke about Crosby's bid on Thursday, Harneet added, the AI-native law firm is about to become just a law firm. Not everyone agrees, though. Ryan Walker, the CEO of GenLegal, wrote, We looked at this and passed. The data belongs to the clients and is not acquirable. The talent who does the work has already left. Instead, we're interviewing Wiles attorneys who want to work with the best engineers in

legal tech to redefine their practice. If Crosby goes through with this purchase, it will be a huge and unprecedented bet on the agentic firm. Certainly, their investors seem to be on board, with Josh Wolf from Lux Capital commenting, wow, Lux family company Crosby are playing to win, and Dick Costello from Index adding, masterstroke from our portfolio company Crosby Legal. Next up, we move over to markets where the cracks are spreading in the AI trade, or at least a clear map of the disagreements between the hyperscalers and investors. Amazon reported 39% annualized growth for AWS, their highest cloud growth in almost five years. However, overall revenue growth came in at 12%, with only $20.7 billion in net income, both of which missed Wall Street forecasts. Meanwhile, Intel's earnings report included data center growth of 78%, and although the company's overall revenue grew in line with expectations at 21%, they forecast a growth slowdown and worsening profit margin due to slowing consumer demand. Amazon stock was down 6% in after-hours trading, while Intel crashed by 23%.

That added to a pretty miserable daytime session for tech, which included a 12% drawdown for Google, which lost half a trillion in market cap and a 1.4% drop for the Nasdaq index overall. Reports from all three companies seem to be telling the same story. AI-related revenues are still surging. Legacy businesses, though, are facing slowdowns. That is a big problem, as their AI businesses are still relatively small, meaning the legacy businesses are subsidizing this AI build-out. Amazon, for example, is now cash flow negative to the tune of $42.4 billion over the past 12 months, while Intel's chip-making foundry division is growing at a 58% pace while making a $1.9 billion operating loss for the quarter. In normal times, these kinds of results could be a sign of an economic slowdown in the consumer sector, particularly with Amazon's retail business and Google's advertising business showing the most weakness. But of course, because we're in the middle of the AI boom, the market pullback is going to be read as the start of an AI bubble burst. That narrative isn't in the financial reports, with AI-related revenue growth still red hot, but you can bet that narrative will find its way into the financial press if tech stocks continue to slide.

Overall, tech companies seem to be in a very difficult position as we head into the end of the year. Their AI businesses are growing incredibly quickly, and that growth supports further investment. During the earnings call, Amazon CEO Andy Jassy said, we believe that AWS can grow north of 40% in each of 2027 and 2028 and be a trillion-dollar revenue business in the next 5 to 10 years. However, the AI businesses are not yet big enough to sustain their own build-out and carry corporate earnings. If the legacy businesses in tech continue to falter it will put even more pressure on AI financing And ultimately whether the strain on AI financing comes because of people being skeptical of the AI trade or because of weakness in the broader economy, the net result could be the same. Analyst Andreas Steno-Larsen summed up a fascinating dynamic happening right now. He wrote, Mag7s are not budging. They keep spending despite markets disliking it. One of the fiercest battles between capital market investors and executives of the tech companies I can recall. Next up, President Trump's use of Grok is back in the news. Democrat leaders are demanding chatbot logs to be preserved for future investigations.

That demand is informed by a Time Magazine profile earlier this month, which mentioned that the President sought advice from Grok on the implications if he ordered the forceful removal of Venezuelan President Nicolas Maduro. According to officials, the piece read, Trump came away thinking Grok was ingenious. A letter to the President earlier this week from Representatives Raskin and Garcia began, we are writing to ask whether you have actually been soliciting advice from Elon Musk's AI chatbot, Grok, in order to help you carry out your official duties as president. If so, the letter asserted, all chatbot logs are considered the property of the United States and must be preserved for posterity. Raskin and Garcia are the ranking members of the House Judiciary and Oversight Committees, respectively. With the midterms looking pretty ugly for the GOP, especially in the House, Raskin and Garcia could become chairs of their committees. Beyond the evidentiary importance of the president's chatbot sessions, Raskin and Garcia said that the entire thing was, in their words, deeply troubling. They wrote, You consulted Grok, but not Congress, on your military deployment to Venezuela, introducing into American foreign policy decision-making a technology that is highly manipulatable and demonstrably error-prone. To the extent any of your administration's myriad blunders,

such as launching a war on Iran, are attributable to over-reliance on AI, the American public deserves to know what advice was rendered, and records of that advice must be preserved. In response to Politico, an administration spokesperson said, The White House preserves all required records in accordance with the law. One quick product flag. If you've ever listened to any of my Operator's Cut style episode, you will have heard me be a huge proponent of voice inputs for AI. And basically at this point, I am pretty much only talking to AI rather than typing. And there is now a new player in that space, with Granola launching a new dictation feature, allowing users to let Granola type for them in any app. Granola says the feature is free and unlimited, and it's now live on Mac and coming soon for iOS and Windows. I think that this sort of voice engagement with AI is just a quintessential part of the shift, and so I would highly encourage you, whatever tool you choose, to go and try it sooner rather than later. Lastly today, a bit of a wild story from the New York Times. The paper of record is claiming that NVIDIA CEO Jensen Huang was the target of Chinese spies. The article claims that a top executive at a Singapore-based data center company, who was actually a Chinese asset, developed a close relationship with Huang.

The reporting centers on an intelligence briefing delivered to President Trump around a year ago, involving Alice Huang, an executive at a company called Megaspeed. Alice had been under surveillance by U.S. intelligence under suspicion of funneling AI chips to China and information to handlers in Beijing. The surveillance revealed that Alice had what sources called a close relationship with Jensen, opening the possibility that she was passing along information about NVIDIA to Beijing. Sources speculated that that could include Trump's thinking on export controls, which Jensen is presumed to have some knowledge of. After receiving the briefing, the president reportedly took no action. Now, the reporting is vague in places, as you might expect when reporting on Intelligence Matters, but it's notable that surveillance of Alice Wong is ongoing, implying that sources saw fit to take the information public and burn an active operation. The article also states that Jensen did not receive a defensive briefing, a warning sometimes given to targets of foreign espionage. That could mean that Jensen was unwittingly being used to gather more information, or maybe even suspected of dual loyalties. An NVIDIA spokesperson said that the company is, quote, confident that if the U.S. government believes that any person is a threat to American industry,

it would let us know. They also argued that chip diversion claims against Megaspeed were, quote, thoroughly investigated and debunked last year. Furthermore, they added that customers like Megaspeed, quote, receive ordinary business information, not sensitive NVIDIA trade secrets, and zero information related to national security. Megaspeed, for their part, said that they are taking the claims extremely seriously and have retained outside counsel to conduct an investigation and are committed to cooperating with the U.S. government. However, they added that the company, quote, cannot respond to a media trial and welcomes formal evidence-based investigation. Now, on the one hand, this is exactly the type of circumstance where it would be very easy for our imaginations to get ahead of reality, especially when the reporting that we have is inherently limited. On the flip side, however, it is the New York Times who's doing the reporting, not some random poster on X. Whatever the truth to it, the fact that it got Laura Loomer and Elizabeth Warren on the same side of an issue is impressive in and of itself. That, however, is going to do it for the headlines. Next up, the main episode. A new study from KPMG in the University of Texas at Austin

found that when people work with AI, similar skills don't guarantee similar outcomes. Researchers studied more than 500 early career professionals and found that the best performers consistently amplified the value of AI by guiding, evaluating, and refining its outputs. These top performers, called AI amplifiers, weren't defined by what they knew alone, but by how they worked with AI. Learn more about what separates AI amplifiers from everyone else at kpmg.com slash us slash AI amplifiers. This episode is brought to you by Cdata. Companies rolling out AI keep hitting the same wall. Token spend is climbing, security is holding agents back, and no one knows how AI is truly being used. The root cause? AI is cut off from the systems and context your business runs on. That's why CData built Connect AI Gateway, one control point between your AI and your systems. Built on the connectivity layer powering Palantir, Google, Salesforce, and hundreds more, each agent gets exactly the data and tools it needs.

It governs down to the record and lets you finally see your org's complete AI usage in one place. Accurate answers, a fraction of the token cost, and a gateway that learns your business with every interaction. See how at cdata.ai. You're in back-to-back meetings all day. you're trying to stay present, but you're also worried you'll forget the decision, the action item, the important next step. This is the scenario that we absolutely live here at the AI Daily Brief, given how fast things move and change and how quickly we're evolving everything that we do. That's where Granola comes in. Granola is an AI-powered notepad for meetings. You jot down rough notes like you always do, and in the background, Granola transcribes and turns them into clear, useful notes when the meeting ends. No bots joining your calls, no distractions, just a clean notepad that helps you focus. During or after the call, you can chat with your notes. Ask Granola to pull out action items, help you negotiate, write a follow-up email, or even coach you using recipes which are pre-made prompts. Once you try it on a first meeting, it is very hard to go without. Head to granola.ai slash ai daily and use code ai daily to get 100% off for three months for new users.

That's granola.ai slash ai daily and code ai daily to get 100% off for new users for three months. Blitzy's understanding of massive codebases unlocks autonomous security fixes, modernization, and new features. So what happens when there's no legacy code at all? Greenfield is supposed to be the easy part. Clean slate, no technical debt. But even Greenfield moves at human speed one sprint at a time. Blitzy changes the unit of work from the developer to the project, autonomously planning, building, testing, and validating entire applications from scratch. Hundreds of thousands of lines of production-ready code. One Blitzy customer stood up a brand new application, 534,000 lines of code, compressing a 65-week roadmap into two weeks. Another shipped an entire application with no front-end engineer. Legacy or Greenfield, the answer is the same. Software at the speed of compute. Build what's next at blitzy.com. That's B-L-I-T-Z-Y dot com. Welcome back to the AI Daily Brief There are a lot of companies that talk about their AI transformation or becoming AI native

And yet most of them, most of us, I should say, because I consider what we do at AIDB and super intelligent, and now how we got here is all part of this, are both using AI to do the same stuff we've always done, but a lot faster and better, and also perhaps increasingly figuring out new things that we couldn't do before. but the outputs and products remain fairly similar. The org structure, although pulled and pushed and twisted in new ways, still would be recognizable to someone from 20 years ago. But what happens when you forget all the old rules of building a startup, take the AI pill, and go completely all in on building with agents? That's what Jacob Sanberry did with his company, Spawn. Over the past two years, he painstakingly figured out how to build a radically AI-first gaming startup. After a lot of trial and error, his 10-man team is now shipping an impressive product built on a huge agentic stack. And this week, Jacob shared his playbook in a blog post entitled, How I Built the Most AI-Pilled Company in the World. Now, before we get into it, a brief explanation of what the product is.

Spawn is a massively multiplayer game creation platform. If you think Metaverse crossed with Lovable crossed with Steam, you're somewhere at least in the right ballpark or playing the right sport. Jacob described it as a massively multiplayer game creation engine where thousands of people can be inside one game, changing its rules, assets, and graphics in real time on a huge range of devices without lag. Developer Simon Willison added, Spawn is a massively multiplayer gaming environment where players can use prompts to the in-game agent Savi to modify the game world. I fired it up on my phone for a few minutes and was able to prompt a castle into existence, then a dragon, then ride the dragon to fly around that castle. Now, I can do a whole separate show about the instantiation of generative worlds in these sort of massive multiplayer games, but that's not what this show is about. In fact, for the purposes of this show, it doesn't really matter all that much what the product is. The point is Jacob's Playbook, and that will work for any company that's at core shipping code and probably for a whole lot that aren't as well. The underlying idea is to take what agents can do now and push it to the absolute limit.

As I mentioned, Spawn is a 10-person startup with a team of agents that they call the fleet. Unlike what most people have been building, Spawn's agent fleet is persistent rather than being spun up and spun down for particular tasks. Each agent has their own name and identity, their own memory, and crucially, ownership of a certain function within the company. There's a team of agents in charge of maintaining the game engine, another for bug hunting, and another for finance. The agents even get Sundays off for some downtime. Spawn spent $3.1 million on tokens last month, 30 times what they spent on payroll. That makes them one of Anthropix's biggest customers. The fleet shipped 77,000 code changes last month, an average of 108 an hour. But more importantly, having an always-on fleet of agents has fundamentally changed how the team thinks about shipping software. Jacob was clear that none of this worked until the launch of Fable. That model has 500x the number of changes they could ship, but it also meant completely rebuilding their operations around Fable's capabilities. Jacob wrote, In general, we've seen that OpenAI's models game their own checks,

Google's models don't stay coherent for very long, and XAI's models just aren't smart enough yet, though they don't do anything cartoonishly bad. Anthropics models stay coherent, make good decisions, and can do the work. In my opinion, the thing that holds them back most is that they won't take enough risk. Now, obviously, Spahn's way of working is extremely rare, to the point that earlier this year, Jacob thought he was using agents the wrong way because his AI bills were so much larger than even his friends were seeing inside big tech. Turned out, however, he wrote, We weren't doing anything wrong, we just think about AI very differently. Most companies run big human teams that use AI, and we run a small human team alongside a big team of agents. To figure out what we were doing differently, I compared notes with a researcher at Anthropic. They talked to a lot of customers, and they said Spawn is the one that works most like Anthropic. So to make this all work, Spawn took a radical approach to building the fleet. As mentioned before, each agent has real system access and clear responsibilities, similar to a human worker. Every day, the agent fleet spawns thousands of helper agents to carry out their granular work. Meanwhile, the human employees use a modified version of Claude Code to work on bigger picture items that need hands-on steering.

The fleet is the infrastructure that allows the game world to operate. When one player changes something in the world with their personal agent, the fleet is responsible for creating a stable build that can update across every player's device while keeping the game world in sync. Understating it greatly, in my opinion, Jacob wrote, this is not a simple problem. And as far as we know, nothing else in the world is even a fraction of it. Functionally, the players are working on a live code base all tied together by the fleet. Now, it was a long and iterative process to get the fleet to where it is today. Jacob explained that most of the time, their problems were solved by removing systems designed to facilitate human work, but which made no sense when applied to agentic systems. The first example they ran across was communication. The agents would spend most of their time telling each other what they were doing. Jacob wrote, Most of the talking in engineering is people saying what they're about to change. Heads up, I'm changing how we store player data, update your work. One message like that touches lots of other work. If every agent tells every other agent, the messages grow much faster than the work. Double the agents and you roughly quadruple the chatter. That is a huge bottleneck. Jacob noted that agents don't actually need

to communicate as much as humans because they're not as specialized as humans. With a human dev team, if your database guy isn't communicating well with your front-end designer, you're going to have a bad time. and the result is collaborating on reviewing a pull request to make sure everyone is clear on the code change. Agents understand the entire system deeply, so the communication doesn't really have a purpose. Earlier in the year, in fact, Spawn killed pull requests and just had the agents work directly in the code base. The result is a system that's completely alien for human developers, but seems to work well for agent-only systems. Changes are made with little concern about breaking things elsewhere in the code. If something breaks due to a change, an agent is Spawn to triage the problem and fix it. Once an hour, Spawn requires the fleet to have everything in line, and test suites are run every 15 minutes to find problems quickly. It turns out that this way of pushing code is surprisingly resilient, because you're not calling humans at 2am to fix a fatal bug. Instead, the fleet is constantly iterating on the code base and fixing problems as they arise. Jacob said that of the 123,000 changes the fleet made over the past year, only about 1,000 were reverted.

The issue that you might have spotted is that this way of working is incredibly resource-intensive. Spawn runs on 5,000 CPUs, 17 terabytes of RAM, and 150 GPUs hosted on AWS, with another 60% in search capacity on standby. The next big problem for Spawn was the human in the loop. Writes Jacob, The agents were waiting for humans to tell them what broke in production. Agents are great at reading the world and inching toward a goal. Many teams still run their agents privately on test machines with copies of the product and expect them to update it and ship it to customers. Then humans relay the faults and bugs back to the agents. That loop is a huge mess and a waste of time. In Spawn, the player's agents automatically detect bugs and file bug reports as they navigate the game world. Those are then fed into the fleet, along with feature requests or issues in the logs. A small model receives the notification, triages the bug or decides if a feature should be added, and passes it to the relevant agent within the fleet. The entire product is designed to be operated and observed by agents, so humans never need to intervene. This makes Spawn somewhat like self-healing software.

When players run into issues, they are detected and fixed automatically. The system design means Spawn operates completely unlike traditional software products. The team considers it an outage if they don't ship changes every hour. And because agents have limits on their memory and context, the system actually functions better when it moves faster. Moving slow or blocking ships is dangerous writes Jacob A system that moves steadily is much more stable than one that frozen and suddenly moves again Getting to a place where we deploy every hour without disrupting our users took a lot of engineering discipline keeping old systems online for a while and weeks of pushing on this as a core goal. He described a typical issue which was detected eight minutes after it affected a player. The cause was named a minute later and the fix was live 11 minutes later. After that, the platform was stable for the next four hours. The next problem was Jacob himself. By October, he realized that his only interaction with the fleet were matters of taste. Should we add this feature? Should we support a new API surface? This was fine at lower scale, but one morning he woke up to 700 requests. He wrote, The agents were finding questions faster than I could answer them.

Jacob also realized that most of these decisions he'd already made hundreds of times before. These changes are too complicated. These ones are fine. Something feels off about this one. Spahn had two years of these decision logs, So they formalized them into a set of principles and trained AI judges to apply them. The judges do one of three things. If they can find a human decision on the record, it applies the same outcome and keeps an audit trail that points to the prior human decision. If the question doesn't make sense, it kicks it back to the agent for revision. And if it's an entirely novel situation, the judge hands the question to a human team member. This system means the human taste still guides the product, but in a much more scalable way. And if you guys ever want to get real weird, ask me why this reminded me of Islamic jurisprudence in the 11th century, and I will go deep. The first AI judges came online on October 6th and made more than 11,000 decisions in the first eight days. Around 7,300 were decided by the judges based on human precedent, around 2,200 were kicked back to the agent, and around 1,800 were passed to a human. Jacob himself only needed to make the call 27 times during that time frame. This unlocked a massive boost to productivity.

Before implementing the judges, the fleet shipped around 1,000 changes per day. After the judges had scaled up, the fleet shipped over 7,000. So with all of these systems in place, the biggest question was what the human team member should do. Jacob wrote, It's a very different job now. Last year and even earlier this year, each of our engineers owned a system and understood it deeply. Today, they're closer to paratroopers. They drop into whatever situation needs them, get familiar fast, and figure out what they can do to unblock the agents. He believes there's two new jobs for humans now the agents are doing the grunt work. The first is taste, but that largely involves playing with the product and figuring out what needs to be changed, rather than micromanaging the development. It also means telling Claude no even if something works. There's also a lot of deleting. Each new model changes what the system can do and makes a lot of the work that came before completely redundant. In June, Jacob looked around at six months of work that encompassed 70% of the engine and perpetually felt like it was just one commit away. He scrapped the entire thing and started over. Eventually, he expects the fleet to take over this duty, but for now, ripping out redundant chunks of the platform is a job for humans.

The other big role remaining for humans is managing entropy. Jacob said this comes in the form of, quote, Asking the models for things they wouldn't have suggested on their own. Ambition is one kind, creative entropy is another. Spawn has found that the models are great at building the next 2% of the platform every week, but they're really bad at recognizing that they've headed in the wrong direction for days at a time and need to take drastic steps to get back on track. On creative entropy, Jacob wrote, The models also won't ask for 100x on their own, so we do. Some of the prompts we've given Claude include things like, Imagine if worlds could be actually unlimited. How do we make a world a thousand times the size of GTA load on the phone in one second? That last prompt was the trigger for the latest build of Spawn, the first one they said that really felt like it delivered on the original vision. And importantly, it never would have happened if the agents were left to their own devices. While Jacob believes that this is the way that his startup needs to operate, he also thinks it's a playbook that will be difficult to replicate. I think big companies will struggle to adopt this, he wrote, because none of it works unless you close the loop, and closing the loop takes risk appetite. You have to be willing to let your agents break production.

He continued, The other mistake is shoving AI into human-shaped boxes. These things aren't human-shaped. Our agents do a lot of work nobody can fully verify, like customer support. Sometimes they're a little bit wrong, but not that wrong, and they recover fast. When we take them off support, our customers get upset. They'd much rather have an AI that answers right away, gives them the clearest information it has, and can do something about the problem on the spot, than a human who says, I'll go look at this, and comes back two days later with a fix. We're not passing our AIs off as human support. We're using them for what they're actually good at. Still, Jacob believes it is entirely worth it. Each new model has made the team work faster and made the system more coherent. Fable was the first big breakthrough that made it functional, but it still had a lot of problems. Fable 5.1 fixed most of those problems, and the team expects that to keep

happening as new models are released. With the right structure, wrote Jacob, every new model makes the whole company significantly better, and that's very hard to catch up with. There are two other things that stand out to me at first glance as other reasons that this is hard to copy as well. The first is that this is the type of system that's hard enough to build when you're building it from scratch, from the ground up, with all of this intention built in from the beginning. It seems to me to be fairly inconducive to modifying an existing system to this new way of working. I would imagine that if I was trying to replicate this, I would do it in some entirely new division or entirely new product rather than trying to reapply this way of working to an old section of the company. The second piece is I wonder how much this relies on the product inherently having an incredibly fast feedback loop built in. The self-healing nature of the system that comes from the actual users of the product with their agents automatically reporting bugs is not the type of thing that you're going to get in many categories of products. Does this sort of truly agent-first approach require that sort of feedback loop? And if it does, are there ways to build that

even into product experiences that don't have a natural spot for it? Still, I think that there are things that other teams can learn, even if we're not going to go as far as Jacob. Five principles and practices that stand out to me include one, giving agents ongoing responsibility. Rather than thinking about them as something to be spun up when the need demands it, figure out how to put them in a perpetual system. Practice two is letting the work carry the context. More agents can create more coordination overhead. A shared and up-to-date record makes it easier to see what has changed, who owns the next step, and what remains unresolved. Third, making the results observable. And this gets a little bit into that feedback loop that I was just discussing. The only way for new agents to get up to speed and start to work in the system that predated them is to have visibility into what has been built and what is currently available, meaning observability becomes incredibly important. The fourth one that I think is highly applicable to lots of companies, turning decisions into precedent. Remember, this is where Jacob woke up with 700 requests and said, wait a second, we have two years of logs of having already made these decisions.

And in this way, decisions end up doing double duty, not only solving the issue at hand, but also building the canon of how issues get solved to make things easier to solve in the future. Lastly, and unsurprisingly, human attention really is on direction. And I think even more than that, understanding that ambition and long horizon views do not come natural to agents is a really important insight that can shape a lot of other work. So that is how the world's most AI-pilled company actually works. Big thank you to Jacob and his team at Spawn for being willing to share publicly how they work. It's going to be fascinating to see as other companies in other areas try to replicate this or do their own version of it, how many of these principles apply across different domains and industries. For now, that is going to do it for today's AI Daily Brief. appreciate you listening or watching as always and until next time peace