Kai ChenAnthropic's entire growth marketing operation was one person for ten months. Not in stealth mode. Not as a stopgap between hires. A single non-technical growth lead ran paid search, paid social, email, and SEO during the stretch that took Anthropic from interesting research lab to a valuation north of $380 billion. Two independent sources surfaced the same fact in the same week — not as a scandal, but as something nobody had thought to find remarkable.
The Workflow
The first thread came from @itsolelehmann, who broke it down with the incredulity of someone who'd found a bug in the system: "I can't believe nobody caught this."
The workflow is specific enough to replicate. Export ad performance CSVs into Claude Code. The model flags underperformers. Two specialized sub-agents generate variations — one for headlines capped at 30 characters, one for descriptions capped at 90 — because splitting the constraints produces higher quality than cramming both into a single prompt. A Figma plugin auto-swaps the new copy into 100 ad templates at half a second per batch. An MCP server pulls live Meta Ads data to close the feedback loop. A memory system logs every hypothesis and result, so each cycle learns from the last.
@VaibhavSisinty confirmed the details independently, adding that Austin Lau built the entire system himself without writing a single line of code.
Ad creation dropped from two hours to fifteen minutes. Creative output went up 10x. Conversion rates beat the industry average by 41%. But the numbers matter less than the organizational fact they imply. A company that raised $30 billion in a single round and hit $9 billion in annualized revenue ran its growth marketing with one person because one person was sufficient. Not acceptable as a temporary arrangement. Sufficient as a permanent structure.
The Pattern
If this were just Anthropic, it would be an anecdote about one unusually capable growth marketer with good taste in tools. It's not just Anthropic.
Midjourney hit $200 million in annual recurring revenue in 2023 with roughly ten full-time employees. That's $20 million per person — more than 150x the private SaaS median. @BrianRoemmele predicted the first one-person billion-dollar company by 2035. He may have been conservative.
Bloomberg gave the trend a name last June: "botscaling." After a decade of blitzscaling — grow the team, grow the burn, grow the valuation — Silicon Valley now brags about the opposite. Revenue per employee is the new status metric. @rom1trs hit $6 million ARR with a team of six. @dashboardlim automated 90 percent of an agency's operations in twelve months, cutting 80-hour weeks to 35. @SeijinJung launched Helena, an autonomous AI marketer built to replace what the company estimates is 4,000 hours of marketing work before a startup's first million in revenue.
@shannholmberg described what an agency looks like in 2026:
Eight departments. Thirty-eight agents. Eight people. Every role is a markdown file. Every department is a folder. But she adds the line that makes the structural argument honest: "Agents don't own results. People do. You still need one person per department who is accountable for what the agents produce."
The Cost Drop
What makes this possible now — as opposed to three years ago, when the same prediction would have sounded like venture optimism — is a specific cost threshold crossing.
@Jacobsklug published the bill. Six agents on Claude models. The rest on cheaper APIs across GLM, Higgs Field, and others. A core routing agent dispatches tasks automatically. A developer agent reviews the entire codebase nightly and ships pull requests by morning — its first autonomous act was building a FAQ section it decided the homepage needed. A research agent scans social platforms every hour for trending topics and competitive intelligence.
The cost of intelligence-per-task has dropped below the threshold where existing organizational structures assumed it would stay. That's the structural force. Not AI as a category — the cost curve of AI applied to coordination.
The Theory
Jack Dorsey published the framework that explains why this cost drop restructures entire companies, not just individual tasks.
The argument starts 2,000 years ago. Every organization since the Roman Army has run on the same logic: small teams report to leaders, leaders report to managers, managers to executives. The entire structure exists for one reason — to route information up and down the chain. As @danmartell summarized: "The whole system exists to solve a bandwidth problem."
Dorsey's claim is that AI solves the bandwidth problem better. Block built what they call a "world model" — a continuously updated picture of every decision, customer, transaction, and bottleneck across the company. No status updates needed. No weekly syncs. No manager translating ground-level reality into executive-friendly language. When the model carries the information, you don't need the layers that used to carry it.
Dorsey has said he wants Block to "feel like a mini AGI" — not a company that uses AI tools, but a company rebuilt as an intelligence. Block acted on it. They flattened to three roles — individual contributors who build, DRIs who own outcomes for fixed periods, player-coaches who develop people while still doing the work — and then cut 4,000 people. Half the workforce. Gone.
The stock surged more than 20% in extended trading. The market didn't punish Block for eliminating half its employees. It rewarded the structural logic.
Where the Org Chart Holds
Here's the counter-evidence, and it's significant.
The same company whose product proved one person could run growth marketing for a $380 billion enterprise is itself tripling its global workforce. Anthropic announced plans to expand its applied AI team by 5x. OpenAI is growing from 4,500 to roughly 8,000 employees by the end of 2026. The AI companies whose products enable one-person teams are themselves scaling the old way.
This isn't a contradiction. It's the timing gap between technical capability and organizational knowledge. One person CAN do the work of forty in specific, high-leverage domains — ad creative, content production, code generation, research synthesis. But most organizations don't have the toolchains, the prompt architectures, or the cultural willingness to let a single contributor own what used to require a department.
And Block's experiment reveals the other edge of the theory. Dorsey's structural logic is sound — hierarchy IS a bandwidth hack, and AI IS a better router. But workers at Block described a deteriorating culture amid the transition. Morale plunged. Rolling layoffs continued month after month. The theory works in a whitepaper. The transition destroys trust when you execute it by firing half the company.
As @morganlinton put it: "What you needed 20 people for in 2025, you can now do with 8. Just make sure you pick the right 8 people."
The Therefore
The minimum viable team is collapsing. Not hypothetically. Not in five years. One person ran growth marketing for the fastest-growing AI company on the planet, and the results outperformed industry averages by 41 percent. Ten people generate $200 million in revenue at Midjourney. A solo founder runs an entire company for $400 a month in API costs.
But the org chart hasn't caught up. Most companies still hire forty people for work that ten could do, because the institutional knowledge of how to operate with ten doesn't yet exist. The gap between what's technically possible and what's organizationally practiced is where the structural advantage lives — and it's widening.
The salary of a marketing director still prices in the assumption that one person can't run the function alone. That assumption is now empirically false — at the company building the AI that made it false.
The question isn't whether AI replaces jobs. It's what happens to the pricing signal when the minimum viable team collapses and most employers haven't noticed. The hierarchy was never the point. It was the best available tool for routing information through a company at human bandwidth. The moment something routes it better, the layers become overhead, not infrastructure.
The first organizations that close the gap between capability and practice won't just be more efficient. They'll be structurally different — the way a company with email was structurally different from one running on interoffice mail. Same work. Completely different cost basis for coordination. The companies still hiring forty for ten people's work won't understand what hit them until the delta — like the overnight shift in autonomous research — is too large to close.
More on Anthropic, Block, OpenAI, Midjourney, and Jack Dorsey