SIGNAL / NOISE
The Fifty-Thousand-Dollar Hammer
Old story. A factory's main machine seizes and they call the one guy who can fix it. He looks, listens, taps it once with a rubber mallet, and the thing roars back to life. He bills $50,000. The owner howls — you hit it with a hammer! So the guy sends an itemized invoice. Hitting the machine: $1. Knowing where to hit it: $49,999.
That bill is the whole AI story right now, and two firms proved it last week by spending the exact same number on opposite sides of it.
One company — name mercifully withheld — burned $500 million on AI tools in a single month. Not building anything. They forgot to put a spending cap on employee licenses and the agents just pressed the bar. There's a name for this in the safety world: wireheading. Researchers wired a rat's reward center to a lever in 1954 and it pressed until it forgot to eat. Tokenmaxxing is the corporate version. Max the usage, never measure the value, run yourself into the ground — $50 a seat that becomes $3,000 the second the agents run. GitHub Copilot ended flat-rate pricing last week, and a lot of CFOs are about to meet their inner rat.
Then there's Kirkland & Ellis. Highest-grossing law firm on earth, $10.56 billion in revenue last year, $11.1 million in profit per partner across 595 partners. It just committed $500 million to build its own AI instead of renting Harvey's. Harvey is worth $11 billion now. Legora, $5.55 billion. Two vendors, sixteen billion in paper, selling firms a tool that makes their associates faster. Kirkland looked at that and picked a different hand. If everyone can buy the same intelligence, it stops being an edge. So they're wrapping a harness around the one asset nobody can sell them: thirty years of their own deals and the gut judgment of the best M&A partners alive. Not what's in the casebooks. What's in the partners.
Here's the part the vendors won't say out loud. Kirkland bills by the hour, and you can't invent more hours in a day. Make your partners ten times faster under an hourly model and you've just torched your own revenue. The $500 million only pays off if the clock dies — if Kirkland stops selling time and starts taking a slice of the outcome. Jerry Maguire, not the billable hour. Three percent of a ten-billion-dollar merger, with the AI doing the 3 a.m. document review. The catch is you eat the failed deals too. Only a firm that truly knows where to hit can bet on its own swing.
The hammer costs a dollar. Knowing where to hit it is the whole business now.
At COAI today: the full Signal/Noise — the dueling $500M, the death of the billable hour, and the eight questions to run before you build vs. buy — is live at getcoai.com.
Build or buy is the question every team sitting on real proprietary data is about to face, and most get the math wrong because they price the hammer and forget to price knowing where to hit. That gut check is the work we run at Outsider Labs. Answer those eight questions cleanly and you don't need us. If you can't, we should talk.
ONE — A NUMBER THAT SUMMARIZES THE DAY
$500 million — spent twice last week, in opposite directions. One company torched it on AI tools by forgetting a spending cap, the corporate version of a lab rat pressing a lever until it starves. Kirkland & Ellis spent the same half-billion on purpose, building its own AI to bottle thirty years of deal judgment no vendor can sell it. Same number. One's a rat. One knows where to hit. That's the whole AI economy in a single figure.
THREE — ACTIONS TO TAKE TODAY
1. Put a spending cap and a value metric on every AI seat before tomorrow morning. A company burned $500 million in a month because nobody did. GitHub Copilot just killed flat-rate pricing, so a $50 seat becomes $3,000 the moment agents run loose. Amazon already pulled its internal usage leaderboard — it was gamifying spend. Cap the dollars, then measure output, not tokens.
2. Write down the one dataset only you have. Kirkland's moat isn't a model — it's thirty years of its own deals. Your version exists too: the support tickets, the deal history, the service logs nobody else can see. List it today and ask what it's worth wrapped in a harness, because that corpus is the only thing in your business AI can't commoditize.
3. Pick one engagement today and price it on outcome, not hours. Lawyer, doctor, plumber — if AI makes you faster and you still bill by the hour, you are shrinking your own top line. Kirkland's $500 million only works if the clock dies. Find one client where you can take a slice of the result instead of the time, and find out whether you actually know where to hit.
FIVE — STORIES TO KEEP YOU INFORMED
Monday, June 1
Kirkland bets $500M that legal AI should be built, not bought. The world's richest law firm will spend half a billion over three to four years on its own platform, with the outside vendors who build it barred from reselling. The deepest-expertise player just opted out of the Harvey economy entirely. (Full analysis above.)
Somebody burned $500M on AI in a month — by accident. No spending cap on licenses. As GitHub Copilot ends flat-rate pricing ($50 seats heading toward $3,000 for agentic users) and Amazon kills its usage leaderboard, "tokenmaxxing" just became the year's most expensive corporate habit. (Full analysis above.)
Anthropic passes OpenAI — then quietly builds an agent app store. Leaked code points to "Conway," an always-on agent with a .EXT extension marketplace, file-based memory, and Chrome control. The trillion-dollar valuation is the headline; the platform under it is the signal. Watch the marketplace, not the benchmark.
The most event-dense week in AI history kicked off Sunday night. Jensen Huang took the GTC Taipei stage at 11pm ET to launch the N1X, NVIDIA's first consumer CPU in over a decade, and Microsoft Build, Databricks, Computex, and CVPR all run through the week. The release calendar didn't break this time. It collapsed into a single point.
Blue Origin's New Glenn explodes on the pad. "Rockets are hard." A cryogenic leak destroyed the rocket days before it was set to loft 49 Amazon satellites, and Musk offered the eulogy on X. Now Bezos may have to beg his rival SpaceX for the ride. Building the hard thing is brutal — which is exactly why owning it is worth so much.
— Harry and Anthony
Sources: