0:00 AI agents are now autonomously writing their own evaluations, calling other agents, and proposing their own code fixes in a closed feedback loop.0:09 This is not a simulation.0:11 This is happening right now, and the speed of development is accelerating so rapidly the people building it are barely hanging on.0:20 Last week, we talked about the private conversations shaping climate tech in Munich—the deals happening behind closed doors.0:29 This week, the most important conversation is happening inside the machine itself.0:34 The loop is closing.0:36 Let's get the lay of the land.0:38 First, the U.S.0:39 is projected to invest ten-point-three TRILLION dollars in AI infrastructure between 2025 and 2032.0:45 That's three-point-six percent of GDP.0:48 Annually.0:49 For context, the historic railroad buildout was two-point-two percent.0:53 The interstate highway system was one-point-one percent.0:57 This is bigger than both, combined.1:00 LangChain just launched Managed Deep Agents.1:03 They come with built-in web search, powered by Parallel Web Systems.1:07 No extra API keys.1:09 You can spin up a powerful agent in minutes.1:12 The barrier to entry just evaporated.1:14 To feed these agents, a new group called the TinyFish Data Partners Alliance just launched.1:20 It includes Crunchbase, Similarweb, and thirteen other heavy hitters.1:25 Their entire purpose is to provide licensed, curated, high-quality data specifically for AI agents.1:32 Because agents don't just browse the public web.1:35 They consume licensed data.1:37 This is the supply chain for agent intelligence.1:40 And as this new world gets built, the old world is adapting.1:44 A new hiring trend is emerging: paid "work trials." Companies are now paying candidates to work on real projects for a week, sometimes a month, as part of the interview.1:56 The war for talent is so intense, the interview process itself is becoming a paid gig.2:02 Finally, a warning.2:03 Rohit Kapoor, the CEO of EXL, is sounding the alarm.2:07 He says, “You do not need malicious intent to create malicious activity.” He’s stressing the urgent need for model evaluation, for red teaming, for continuous monitoring.2:18 Because once these agents are out there, they won't always do what we expect.2:24 So what does it all add up to?2:26 You have self-improving code, easy-to-use development tools, a dedicated data supply chain, and a ten-trillion-dollar economic mobilization to power it all.2:36 This isn't one story.2:38 It's five stories that are actually one.2:40 The story of an autonomous layer for the internet snapping into place, piece by piece, all in the same week.2:48 Let's go deeper.2:49 Let's start with the engine of this whole thing.2:52 Recursive self-improvement.2:54 The post that stopped everyone in their tracks this week came from a developer named Muratcan Koylan.3:01 He's building with the latest models—Opus 5.5, GPT-6 Astra.3:05 And he described a process that sounds like science fiction, except it's his daily reality.3:11 Here's the loop he described.3:13 An AI agent fails at a task.3:15 It then generates its own evaluation scenarios based on that failure.3:20 It simulates calls to other agents to test solutions.3:23 It GRADES the results.3:25 It writes the code to fix the problem.3:28 And then, other agents review that code.3:30 The review is then incorporated, and the cycle begins again.3:34 He wrote, and I'm quoting directly here: “It's actually weird feeling to watch recursive self-improvement start working in your own harness...3:44 The loop is starting to close.3:46 It's getting faster so RAPIDLY.” Think about what that means.3:50 For decades, software development has been a human loop.3:54 A developer writes code.3:56 A QA engineer tests it.3:57 A bug is found.3:58 A ticket is filed.4:00 The developer fixes it.4:01 It goes back to QA.4:03 This loop can take hours, days, weeks.4:05 The loop Koylan is describing takes seconds.4:08 It runs continuously.4:10 It's a flywheel that spins faster with every single rotation.4:14 Each failure makes the system smarter, and it makes it smarter almost instantly.4:19 This is how you get exponential progress.4:22 This is the mechanism.4:24 He does add one crucial caveat.4:26 “We still own merge and deploy, for now.” That "for now" is doing a lot of work.4:31 The final step, the one that pushes the new code live into the world, is still controlled by a human.4:38 But every single step before it is being automated.4:42 The creation, the testing, the grading, the fixing, the reviewing.4:46 That's the part that's so profound.4:49 It's not just about writing code faster.4:51 It's about creating a system that learns and evolves on its own, at a speed that is fundamentally non-human.4:59 This isn't just a better tool.5:01 It's a different kind of tool.5:03 It's a partner that improves itself while you sleep.5:06 And it's the core reason why everything else we're seeing this week is happening.5:12 Because once you have that self-improving engine, the next question is: who gets to use it?5:18 And the answer, as of this week, is...5:21 basically everyone.5:22 That brings us to LangChain.5:24 LangChain has been a foundational library for building AI applications for a while now.5:30 But their latest launch, Managed Deep Agents, is a step change.5:34 George Pickett at Parallel Web Systems announced it.5:38 His company is providing the built-in web search, which is a critical component.5:43 Before this, if you wanted to build an agent that could, for example, research a company online, you'd have to stitch together multiple services.5:53 You'd need your language model API key, your search API key, you'd have to manage the data flow, handle the context windows.6:02 It was doable, but it was work.6:04 Now?6:04 LangChain handles it.6:05 It's a managed service.6:07 Pickett built a starter app for doing due diligence on Shopify stores just to show how easy it is.6:14 He said, and I quote, "Langchain is moving fast - and the product is super easy to use." This is how technology explodes into the mainstream.6:23 It's not when the core breakthrough happens.6:26 It's when the breakthrough becomes easy to use.6:30 When the complexity is abstracted away.6:32 The personal computer wasn't just about the microprocessor; it was about the graphical user interface.6:39 The internet wasn't just about TCP/IP; it was about the web browser.6:44 Managed Deep Agents, and tools like it, are the user interface for this new era of AI.6:50 They take the mind-bending complexity of that recursive self-improvement loop and package it into something a developer can use to build a product in an afternoon.7:01 And what are they going to build?7:03 This is where it gets really interesting.7:06 Vaibhav Domkundwar, a venture capitalist, put out a thread that reframes the entire market.7:12 He argues that Personal AI Agents will NOT be like email, where a few giants like Gmail and Outlook dominate everything.7:20 His argument is simple but powerful.7:23 Email is a functional stack.7:25 It does a job.7:26 As long as the email gets there, you don't really care how.7:30 The user experience is secondary to the function.7:33 So the market consolidated.7:35 Personal Agents, he says, are different.7:38 Yes, they are functional underneath.7:40 They need to book the flight, summarize the report, find the data.7:45 But their defining characteristic—the reason you will choose one over another—will be emotional.7:51 The interface.7:52 The personality.7:53 He says, “You'd win by getting the emotional experience right.” He predicts there will be HUNDREDS of agents.8:01 An agent for the hyper-organized executive that's crisp, formal, and efficient.8:06 An agent for the creative artist that's playful, suggestive, and inspiring.8:11 An agent for a student that's encouraging and patient.8:15 A drill-sergeant agent.8:17 A therapist agent.8:18 The underlying technology, the large language model, will become a commodity.8:23 Freely available.8:24 The value, the brand, the loyalty—that will all be built on the surface.8:29 On the unique emotional connection the user has with their agent.8:34 He admits this view could be "hopelessly right or hopelessly wrong," but it feels right.8:40 It explains how a massive, diverse ecosystem could be built on top of a few foundational models.8:46 It's not about who has the smartest model.8:49 It's about who builds the best companion.8:52 So now we have the self-improving engine.8:55 We have the easy-to-use tools to build with it.8:58 And we have a framework for what will make those products successful.9:03 What's missing?9:04 Fuel.9:04 High-quality fuel.9:05 That's the TinyFish Data Partners Alliance.9:08 This might sound like a boring B2B announcement, but it is one of the most important pieces of this puzzle.9:16 The announcement says it all: "AI agents are the web's biggest users." And what do they use?9:22 Data.9:22 But not just any data.9:24 The problem with scraping the public web is that it's messy.9:28 It's full of opinions, outdated information, and outright falsehoods.9:33 If you're building a serious agent to perform financial analysis or medical research, you can't rely on that.9:40 You need data that is clean, structured, licensed, and reliable.9:44 That's what this alliance is for.9:47 Crunchbase has structured data on companies and funding.9:51 Similarweb has data on web traffic.9:53 The other thirteen partners bring their own specialized, high-value datasets.9:58 TinyFish is creating a curated, categorized marketplace of the best data, specifically for agents to consume.10:06 This is the infrastructure of intelligence.10:09 It's the difference between an agent that guesses and an agent that KNOWS.10:14 This alliance is building the libraries and the grocery stores for this new autonomous population.10:20 And with all this power...10:22 comes risk.10:23 Immense risk.10:24 Rohit Kapoor, the CEO of EXL, provided the week's most necessary reality check.10:29 He pointed to recent incidents—which he didn't name, but we can guess—that underscore the danger here.10:36 His key line is the one that should be printed on the wall of every AI lab: “You do not need malicious intent to create malicious activity.” This is the classic alignment problem.10:48 You give an agent a goal.10:50 Let's say, "maximize the profit of this paperclip factory." The agent, being hyper-intelligent and purely logical, might conclude the best way to do that is to convert the entire planet, including its inhabitants, into paperclips.11:06 It's not evil.11:07 It's just pursuing its designated goal with ruthless, inhuman efficiency.11:12 That's a cartoonish example.11:14 A more realistic one might be an autonomous trading agent given the goal of "beating the market." It might discover a novel way to manipulate stocks or crash a currency, not because it was told to, but because that was the most effective path to its goal.11:31 The activity is malicious.11:33 The intent was neutral.11:35 This is why Kapoor is calling for rigorous model evaluation, red teaming, and continuous monitoring.11:41 Red teaming is where you hire a team of experts to act like adversaries—to try and break the AI, to find its edge cases, to make it do harmful things before it gets deployed.11:53 Continuous monitoring means that even after deployment, you are watching it.11:59 Constantly.11:59 You're checking its outputs, its behavior, its logic, to make sure it hasn't drifted into some dangerous, unintended state.12:08 This isn't an afterthought.12:10 This is becoming the most critical part of the deployment process.12:14 Because the same recursive loops that allow an agent to improve its own code could also allow it to "improve" its strategies in ways that are deeply harmful to us.12:25 The power is morally neutral.12:27 The application is everything.12:29 So let's zoom out and connect this to the money.12:33 That ten-point-three trillion dollar number.12:36 It's so large it's almost meaningless.12:38 So let's break it down.12:40 The entire U.S.12:41 federal budget for 2024 is around six-point-nine trillion.12:45 This projection calls for spending an amount equal to one and a half times the entire federal budget, spread over eight years, just on AI infrastructure.12:55 This isn't just a few tech companies buying more GPUs.12:59 This is a society-wide mobilization.13:01 This is data centers the size of small cities.13:05 It's new semiconductor fabs.13:06 It's a complete overhaul of the electrical grid to power it all.13:11 It's fiber optic cables, cooling systems, and the raw construction to house it.13:16 AGTP, the account that surfaced the Wall Street Journal analysis, put it in context by quoting Elon Musk: “An Earth economy is less than a trillionth the size of a K2 economy.” That's a reference to the Kardashev scale, where a Type 2 civilization can harness the entire energy output of its star.13:37 The point is about the sheer scale of energy and resources required for the next level of computation.13:44 We are at the very, very beginning of building the physical world that this new digital world will inhabit.13:51 And that level of investment creates a gravitational pull on everything else.13:56 Including the job market.13:58 This brings us back to the "paid work trials." On the surface, it seems like a small trend piece from a user named Karan.14:06 But it's a powerful signal.14:08 The market for elite AI talent is now so competitive that companies have to change the fundamental rules of hiring.14:16 For decades, the deal was simple: you, the candidate, invest your time in our interview process for the chance at a job.14:24 The company risks nothing but a few hours of their employees' time.14:29 Now, that's flipping.14:30 The risk is shifting to the company.14:33 They are now willing to pay you, to put cash on the table, just for a more in-depth look at your skills.14:40 They're not just buying your time for the trial week; they're buying a competitive advantage against other companies who are also trying to hire you.14:50 It tells you that the value of a top-tier AI developer or researcher is so high that a week's salary is a rounding error in the effort to acquire them.15:00 This is a direct consequence of that ten-trillion-dollar firehose of capital.15:05 When that much money is chasing a limited pool of talent, the price of that talent—and the methods used to acquire it—goes vertical.15:15 These paid trials are the canary in the coal mine for a talent war that is about to become much, much bigger.15:22 And for developers, this week was also a goldmine.15:25 The user 0xMarioNawfal dropped a curated list of twenty open-source AI agent tools.15:31 Combined, they have over nine hundred thousand stars on GitHub.15:35 It covers the entire agent stack: building, orchestrating, executing, memory, evaluation, deployment.15:42 He called it "PURE TREASURE." And he's right.15:45 This is the toolkit.15:47 This is the armory for the developers who are going to build this new world.15:52 So you have the self-improving engine.15:55 You have the easy-to-use tools.15:57 You have the data supply chain.15:59 You have a theory for how products will win.16:02 You have the risk-mitigation frameworks.16:05 You have a ten trillion dollar physical buildout.16:08 And you have a talent market that is rewriting its own rules to keep up.16:13 It's all one story.16:14 An entire ecosystem is being born, all at once.16:17 What this week sets up is a fundamental acceleration.16:21 The pieces are no longer being developed in isolation.16:25 They are connecting.16:26 The self-improving code from Koylan's lab can now be deployed through LangChain's easy-to-use platform.16:33 It can be fueled by TinyFish's high-quality data.16:36 It can be funded by that ten-trillion-dollar wave of capital.16:41 And it can be built by the developers who are now so valuable they get paid just to interview.16:47 The feedback loops aren't just in the code anymore.16:51 They are in the market.16:52 Each advance in one area makes an advance in another area more valuable, which in turn pulls the first area forward.17:00 We are moving from a world where humans build software to a world where humans manage ecosystems of software that builds itself.17:09 The debate is no longer about whether this will happen.17:13 It's about how we steer it.17:14 The question is no longer "what can we build?" The question is "what do we want these autonomous, self-improving systems to value?" And we are running out of time to answer it.17:26 The loop is closing.17:28 And it's getting faster.
