ONE — A NUMBER THAT SUMMARIZES THE DAY
9 → 90 — pre-AI, the productivity gap between the worst lawyer in a firm and the best lawyer was nine points. The folk-wisdom 10X ratio. Bearable on a P&L, sheltered by hierarchy. Post-AI it became ninety points — same ratio, ten times the absolute spread — and the shelter became a tax on the people doing the work. This afternoon the CEO of ClickUp cut 22% of his staff while calling the business "the strongest it's ever been," and routed the savings into $1M cash bands for the survivors. The institution era is over. Pick a song.
THREE — ACTIONS TO TAKE TODAY
Stop writing job descriptions for the middle. Look at your headcount plan for the next two quarters. If any role you're hiring for is "general operator," "associate," "coordinator," or "manager of [thing]" — kill it tonight. AI just deleted the middle of the org chart. Replace those roles with one of three role-types: builder, agent manager, or front-liner with customer-contact time. That's the new chart. Build it on paper this week.
Run the 9-vs-90 math on your team. Pre-AI, the best engineer / PM / analyst on your team was roughly 10X the worst. A 9-point absolute gap. Bearable on a P&L. Post-AI, that gap is now 90 points because the top performer can wield agents and the bottom one cannot. ClickUp's Zeb Evans published the proof today: "AI makes the best engineers wildly more productive, and everyone else using AI slows these engineers down." If you have not had this conversation with your CFO this month, you are operating with stale numbers.
Pick one of three exits and start moving today. Build — start a company while the barrier to entry is zero. Automate — eat your own job with AI before someone else does it, and become the irreplaceable Agent Manager. Touch — be the human in the customer meeting that no agent can replace yet. The fourth exit — be sheltered by tenure — has been closed. There is no waiting it out.
Today's actions are Outsider Labs in microcosm. The audit business we're building with our partner is exactly the second exit: the outside firm that goes into your company and identifies where AI is going to break your org chart, where the bottleneck is forming, and how to redistribute the work so your 100X people are not drowning in the AI slop of the people around them. Most companies cannot do this from the inside. It is too close to the bone.
FIVE — STORIES TO KEEP YOU INFORMED
Thursday, May 21
1. ClickUp cut 22% while calling the business "the strongest it's ever been." Zeb Evans's 1,400-word post this afternoon is the most honest corporate communication of the year. Twenty-two percent gone. Not for cost reasons. Most savings flow back into the survivors as $1M cash bands. Evans names three roles that survive — Builders, Agent Managers, Front-Liners — and pronounces "the great reckoning of AI coding" on every company celebrating 500% more pull requests with no customer outcome to show for them. (Full analysis below.)
2. Anthropic projects its first profitable quarter — at five years old. Bloomberg leaked the Q2 numbers today: $10.9B in revenue (more than 2x Q1's $4.8B), $559M in operating profit, $900B valuation talks live. Five years. The previous firm to outpace Anthropic's revenue per employee was 222 years old this morning and is now a morgue. October IPO is on the table. (Full analysis below.)
3. Bristol-Myers Squibb put Claude inside the FDA submission workflow for 30,000 employees. The Yahoo Finance writeup reads like a vendor press release. The real story: pharma is the hardest enterprise sale in tech because of GxP and regulatory governance. BMS just crossed the firewall — research, clinical, manufacturing, commercial, and the regulatory submission machine. Every regulated-industry CIO just lost their "we're waiting for it to mature" line.
4. OpenAI's reasoning model autonomously disproved an 80-year-old Erdős conjecture, verified by Tim Gowers. The proof, announced this week, found a new family of constructions that beat the long-assumed square-grid arrangement on the planar unit distance problem. Fields Medalist Tim Gowers and Princeton's Will Sawin signed off. This is the first time a prominent open problem in a mathematical subfield has been solved autonomously by AI. Two days before OpenAI's rumored confidential S-1. The timing is the pitch.
5. California signed the first state-level executive order on AI labor displacement subsidies. Governor Newsom's order, signed this afternoon, directs state agencies to study policies including subsidies for companies that do NOT replace workers with AI. This is the first concrete state-level response to the institution era ending. The 22% who got cut at ClickUp will eventually vote. Pay attention to which other Democratic states follow.
In every era, one man writes. Another waits. AI just made waiting fatal.
SEVEN — SIGNAL / NOISE
The 222-year roof just came off.
Earlier this afternoon, Zeb Evans — founder and CEO of ClickUp — posted a 1,400-word announcement on X. He cut 22% of his staff this morning. The business, he said, is the strongest it has ever been. The savings are not going to shareholders. They are flowing back to the survivors as million-dollar cash compensation bands. He calls it the 100X organization. Read carefully it is the most important corporate communication of the year — and the only honest one.
The thesis, in one sentence: AI does not lift all boats. The common narrative is wrong. "AI makes the best engineers wildly more productive," Evans wrote, "and everyone else using AI slows these engineers down." That sentence is the obituary of the institution era of American knowledge work.
Here is the math. Pre-AI, the productivity gap between the worst lawyer in a firm and the best lawyer was roughly nine points — a 10X ratio compressed onto a P&L that hierarchy could absorb. The 1X person got paid because the 10X person couldn't replicate themselves. Capacity needed bodies. Tenure, partnership tracks, corporate ladders, MBA pipelines, the consulting up-or-out — every institutional structure in the American economy was designed for that 9-point gap.
Post-AI, the gap became 90 points. Same 10X ratio. Ten times the absolute spread. The 10X person, now wielding agents, no longer needs the 1X person to make capacity — Codex finishes the ticket while she sleeps. The 1X person using the same tools is not a 2X person. They are a net drag, because the 10X person now spends her time cleaning up the AI slop of the 1X person instead of directing her own agents. The shelter became a tax.
Today's other data points all tell the same story. Anthropic at five years old is projecting its first profitable quarter — $10.9B in revenue, $559M in operating profit, on track for a $900B valuation. Bristol-Myers Squibb deployed Claude to 30,000 employees. Hark closed a $700M Series A — Brett Adcock's second company in three years. And OpenAI's reasoning model autonomously disproved an 80-year-old Erdős conjecture, verified by Fields Medalist Tim Gowers. The 100X mathematician is not a person. It is a model.
Three exits remain. Build — start a company while the barrier to entry is zero. Automate — eat your own job with AI before someone else does, and become the irreplaceable Agent Manager. Touch — be the human in the customer meeting that no agent can replace yet. There is no fourth exit. The 9-point gap was the fourth exit. It has been closed.
I sat in a law firm this morning on the 32nd floor of 120 Broadway, signing a refi in pen, while a marble bust of a man I didn't recognize watched me from the shelf. The man was Thomas Addis Emmet — an Irish revolutionary who arrived in November 1804 to fill the void left when Alexander Hamilton died on Burr's pistol at Weehawken. The firm he founded — Emmet, Marvin & Martin — ran through the entire history of American capitalism. It argued Gibbons v. Ogden before John Marshall's Supreme Court. Franklin Delano Roosevelt was a partner there. It lasted 222 years. I didn't know whose firm I was in. The bust was Rosebud. Most executives reading this are sitting under their own version of the same bust, in their own version of the same firm, signing their own version of the same paperwork. They do not yet know that the duel they thought their predecessors had ended is back. They will know by the end of June.
The future will be won by the builders, not the readers/talkers. AI exposes capitalism, exposes agency, exposes a willingness to bet on yourself, on your own intellect, on solving problems. You want to just live out your life and pay your bills?
You are dead.
— Harry and Anthony
Sources: