Preface: The Right Side of the Line
In the time it took the markets to react, a trillion dollars in software value evaporated. Not because anything broke. Because something finally registered: a new class of workers had arrived, and the market priced the old class accordingly.
This book is the map for the people on the other side of that repricing. The ones who will build, deploy, and own those workers, instead of being replaced by them.
Teaching Aid
What Just Happened
On February 4, 2026, global software stocks suffered their worst stretch since the 2022 rate-hike selloff. Nearly $1 trillion in market value was erased from the software and services sector over six consecutive sessions. Traders called it the "SaaSpocalypse."
The trigger: Anthropic released eleven open-source plugins for Claude Cowork (its agentic productivity platform), targeting legal, finance, sales, marketing, and data analysis. One plugin could triage NDAs, track compliance, and review contracts. The market's response was immediate. Thomson Reuters fell 16%. RELX dropped 14%. Salesforce and ServiceNow each shed roughly 7%. A single legal plugin handling NDA triage and compliance tracking wiped $285 billion off software, legal tech, and professional services in one trading session.
The message was unmistakable. Autonomous agents can now perform the complex professional work that justified $200/month software subscriptions. The era of paying for a "seat" so a human can click buttons in enterprise software is ending.
By March, the verdict was official. Time Magazine called Anthropic "the most disruptive company in the world." The share of U.S. companies paying for Claude tools hit 20% in January, up from 4% a year earlier.
This was not a product launch. It was a market reclassification of where work, and the value of work, happens.
Then It Got Worse
Three weeks later, on Monday, February 23, a 7,000-word hypothetical from Citrini Research went viral, and the Dow dropped 800 points in a single session. The piece was not a prediction. It was a thought experiment dated June 2028, exploring what happens when AI agents displace white-collar knowledge workers at scale: mass unemployment, software-backed loan defaults, financial contagion. The market treated the thought experiment as a trading signal.
Datadog, CrowdStrike, and Zscaler each plunged more than 9%. IBM fell 13%, its worst single-day performance since 2000. American Express, KKR, and Blackstone, all named in the report, tumbled.
The Citrini line that captured the moment:
"For the entirety of modern economic history, human intelligence has been the scarce input. We are now experiencing the unwind of that premium."
A premium that unwinds has to flow somewhere. The next section is who the market thinks it flows to.
Then It Got Bigger
Seven weeks after the Citrini selloff, the repricing came back to Anthropic itself, this time as a bid, not a sell. In mid-April 2026, Bloomberg reported venture capitalists offering term sheets at $800 billion, more than double the $380B post-money of February's Series G, and within striking distance of OpenAI's $852B. Per Sacra, Anthropic's annualized run-rate hit $30 billion in March, up roughly 1,400% year-over-year.
The comparison that matters is not with OpenAI. It is with the incumbents being disrupted.
At $800B, Anthropic is worth 3.2× the combined market capitalization of India's six largest IT services companies (TCS, Infosys, HCLTech, Wipro, LTIMindtree, and Tech Mahindra), which together employ roughly 1.9 million people and generated around $100 billion in combined revenue last year. Anthropic has approximately 1,000 employees.
The market is not saying Anthropic is a more valuable company than the Indian IT services industry. It is saying Anthropic sits on the right side of the repricing, and those 1.9 million jobs sit on the wrong side, because the software categories those workers build, integrate, and maintain are the categories being replaced. That side of the line is what this book teaches you to occupy.
Then It Got Capable
The repricing was not only about revenue. In early April, Anthropic released Claude Mythos through an initiative called Project Glasswing. The Mythos preview autonomously discovered thousands of zero-day vulnerabilities across every major operating system and web browser, including a 27-year-old OpenBSD bug and a 17-year-old FreeBSD remote code execution flaw. It scored 73% on expert-level capture-the-flag cybersecurity tasks and was the first model to solve a 32-step simulated corporate network attack end-to-end.
Anthropic deemed Mythos too dangerous for broad release and offered access only to select security partners. Treasury Secretary Bessent and Fed Chair Powell reportedly convened Wall Street leadership behind closed doors to discuss the systemic risks.
Mythos is what the rest of this book is about, made concrete: a single AI worker performing the work of an elite human security team, faster, at greater coverage, and continuously. The mid-April bids that took Anthropic from $380B to $800B were not just pricing in revenue acceleration. They were pricing in what one AI worker can now do that an elite human team cannot.
That is encoded domain expertise. That is the factory output.
Then It Got Industrial
On April 9, five days before the $800B bid landed, OpenAI told its own shareholders Anthropic was operating on a "meaningfully smaller curve" — compute-constrained at 7–8 GW through 2027 against OpenAI's planned 30 GW by 2030. Three weeks later, the bid moved again.
In early May 2026, the Financial Times reported Anthropic was weighing a $50 billion raise at ~$900 billion pre-money, post-money within striking distance of $1 trillion. Annualized revenue had moved from $30B in March to approaching $45B by early May, a fivefold jump in five months.
The more telling shift was on the cost side. Inside the same window, Anthropic committed $200 billion over five years to Google Cloud, more than 40% of Google's disclosed cloud revenue backlog. It took every GPU at SpaceX's Colossus 1, 220,000+ NVIDIA chips and 300+ MW of capacity, online within the month. It signed a 5GW agreement with Amazon, locked in 5GW of Google–Broadcom TPU capacity for 2027, took $30B of Azure capacity through Microsoft and Nvidia, and committed to a $50B U.S. AI infrastructure partnership with Fluidstack. Active discussions with SpaceX cover multiple gigawatts of orbital compute.
The funding flowed both ways. On April 24, Google formally committed up to $40 billion of equity into Anthropic — $10 billion deployed that day, up to $30 billion more contingent on performance milestones. The structure was circular by design: Google's investment helped fund the chips Anthropic had just committed to buy back through the Google Cloud deal. The hyperscalers were no longer just selling capacity to Anthropic; they were pre-financing the workforce that runs on top of it.
The shape of the company changed. Anthropic is no longer being valued like an AI lab. It is the year's largest single buyer of frontier compute, electricity, and data-center capacity. The thing being manufactured at industrial scale is not the model. It is the workforce that runs on top of it.
That is the side of the line this book maps. Not the chatbot. Not the seat. The factory.
The Early Tell, in Retrospect
None of this should have been a surprise. In June 2023, Thomson Reuters paid $650 million for Casetext, whose CoCounsel AI assistant scored 97% on complex legal evaluations. The acquisition was not for the technology. It was for encoded legal expertise: the ability to do substantive legal work that previously required expensive human professionals. CoCounsel was the proof of concept three years before SaaSpocalypse made the pattern market-wide; the $800B bid is the price of being on the manufacturing side.
What the Market Is Pricing
Strip away the volatility, and five signals come through.
The SaaSpocalypse proved that autonomous agents can do the professional work that seat-based software was charging for.
Citrini proved that the market expects the disruption to extend, well beyond software, to the broader category of knowledge work.
The trillion-dollar bid proved that the market is willing to pay, today, for the company manufacturing the agents that do it.
Mythos proved that AI workers can now do work that no team of humans can match — not faster, not cheaper, better.
The compute buildout — $200B to Google, every GPU at Colossus 1, 5GW with Amazon, multi-gigawatt orbital plans with SpaceX — proved that the workforce is being manufactured at the scale of national infrastructure, not shipped as a product.
Five different events. One coherent thesis: value is shifting from those who use software to those who own the agents that run on top of it.
What This Means for You
If your job, or the value you sell to clients, depends on humans navigating legacy software, you are being disrupted. Every per-seat contract signed in 2026 has an implied expiration date.
The same shift that makes seats worthless makes something else extraordinarily valuable: encoded domain expertise. Whatever know-how you carry, about contracts, audits, sales motions, supply chains, compliance, patient triage, or classroom assessment, is exactly what AI workers need to be useful. The companies that win the next decade will take that expertise out of human heads and encode it into agents that perform the work continuously, governably, and at scale.
That is what this book teaches.
We call those workers Digital FTEs: role-based, supervised, spec-driven AI agents performing real work inside real organizations. They are not chatbots. They are not demos. They are the new factor of production. The thesis calls them AI Workers; same workers, business-facing register.
Companies built around them, where the workforce is mostly digital and the product is whatever that workforce ships, are AI-Native Companies. The discipline that manufactures the workforce is the Agent Factory: a spec-driven, human-supervised, agent-tool-powered practice. Not a product you buy. A practice you adopt. This book is its canonical source.
If the SaaSpocalypse priced the death of seat-based work, this book is what to do with the room that opens up.
The Build Is No Longer the Bottleneck
The primary interface for constructing AI workers is now natural language: English, Urdu, Spanish, whatever language you think in. You describe the job; the agent assembles the solution. Domain experts without traditional programming backgrounds are shipping production AI workers this way. "Agentic coding" is the discipline that makes it possible; Door 03 below is the fastest hands-on entry into it.
Start Here: Four Doors In
You don't need a PhD. You don't need years of experience. You need the right map, and four free, open resources from The AI Agent Factory that walk you through it, step by step.
🏛️ The Thesis
How AI Agents, AI Workers, and AI-Native Companies fit together, in plain language.
Read the Thesis →🧩 AI Worker Catalog
The real jobs AI Workers are doing today across Sales, Finance, Support, Engineering, HR, and Legal.
Browse the Catalog →⚡ Agentic Coding Crash Course
Hands-on first build. Simple. Practical. You will be coding agents before you know it.
Start the Crash Course →🧠 Part 0: Thinking is the Curriculum
The book's actual starting point. Eleven chapters of thinking skills before tools enter the picture.
Start with Part 0 →These four pages are the entry into a larger curriculum. The book itself is structured into eleven parts, taking you from foundational thinking skills (Part 0), through general-purpose agents (Part 1), to building enterprise-grade Digital FTEs (Parts 6–8), and on to deploying realtime voice and TypeScript agents (Parts 9–10).
You don't have to read it in order. But you should start with the four doors.
Welcome
The market has been telling you what is coming. This book is what to do about it.
Open the first door. That is all it takes.
Flashcards Study Aid
Last updated: May 2026.