Everyone keeps trying to make the model smarter. That is not the enterprise AI bottleneck. The bottleneck is that most companies have no machine-readable operating memory. The customer nuance lives in someone's head. The exception lives in a Slack thread. The decision rule lives in last quarter's meeting. The next step lives in an inbox. The reason the work matters lives nowhere. Then we drop an AI model into the middle and act surprised when it gives generic output. The model is not the missing layer. The missing layer is company context structured around action: who owns what, what happened before, what good looks like, what changed, and where the handoff goes next.

Read the next layoff memo like an operating X-ray. The company cuts 9%. The memo says efficiency. The earnings call says margin discipline. The job board still has AI operations roles open. The VP layer stays untouched. The support queue gets "automated." The customer still waits 36 hours for an answer. That is not an AI strategy. That is payroll reduction wearing an AI costume. If the same manager layer stays, the same handoffs stay. If the same approvals stay, the same delays stay.

The meeting ends with the right answer on the table. Everyone knows it. The account lead knows it. The operator closest to the customer knows it. The founder can feel it in the room. Then the decision goes to a manager who won the job five years ago and stopped being right three years ago. Now the company waits. Not because the work is hard. Not because the team is lazy. Because authority got frozen in the org chart after reality moved. AI makes this uglier, not prettier. When execution gets cheap, stale authority becomes the bottleneck. The person closest to the truth can move faster than the person allowed to approve it. Watch the handoff there. That is where the work dies: between the person who knows and the person still allowed to decide.

Confession: going to make my ai write a 600 page novel where a guy just sits there the entire book.

Every company says the problem is strategy. It almost never is. The sales call happened. The note was written. The follow-up was promised. The ops person was tagged. The founder assumed someone owned it. The customer waited. Can you see the handoff? That's where the work died. Not in the meeting. Not in the CRM. Not in the employee's intentions. In the space between "we discussed it" and "someone actually owns the next move." Founders paying attention right now: the next operating advantage is not a smarter tool. It's a layer that keeps the handoff from disappearing.

Just watched backrooms

I use auto-pencil in Sudoku. Zero shame. The puzzle is not "can I manually write 83 tiny candidate numbers without missing one." The puzzle is the logic. That's how I think about AI at work. Spreadsheets replaced rooms full of human calculators. They did not make business less analytical. They moved people up a layer. AI is doing the same thing to the clerical layer of knowledge work. The point is not to avoid thinking. The point is to stop burning your best thinking on pencil marks.

Most 2026 AI advice is "learn prompting." Buy a prompt library. Take a course. Pay for the secret prompt pack. None of it is wrong. None of it is the game either. Prompting is a typing skill. Operating leverage is an architecture decision. Six turnarounds taught me the same thing every time: the win is not better inputs into the same broken loop. The win is a loop that runs whether you touch it or not. Founders polishing prompts for a year are doing motion. The ones rebuilding the operating layer are doing work.

The internet opened every piece of human knowledge to everyone. 20 years later, most people use it to argue on Facebook. The tools were free. The arbitrage was sitting there. Most people never touched it. AI is the internet all over again. Same shape, same sorting. There will be players, pretenders, and people who never notice it happened. The visionaries and founders paying attention right now are not the majority. They never are.

What if I told you internal meetings aren't work?

Orgs keep asking: how do i put AI in my business. Wrong question. The right one is: which operator decision am i making by hand every week that AI could just make. The first question gives you a Slack bot nobody opens. The second gives you back ten hours and three blind spots you didn't know you had. Tooling chases the first. Operators chase the second.

Most services founders think they need better employees. They need fewer layers between them and the work. Every stall i've watched between $1M and $10M ARR follows the same shape: founder hires another VP, adds a process, ends up one more step from the actual day. The cure was never another hire. It was visibility. The agencies that broke through their stall didn't grow the org chart. They flattened it, and the founder started seeing live work again.

Block is quietly doing something most companies aren't ready to admit out loud: using AI to eliminate middle management. Not reduce it. Eliminate it. Here's why it matters. Every org layer between a founder and reality is a lie compression function. Each layer smooths, softens, strategically omits. By the time information reaches the top, it's a story, not truth. Pre-AI, only workaholics like Musk could beat it, by going straight to the source at 2am. AI changes the math. A founder with the right operating layer can now see reality directly, without middle-management translation. This is not "productivity." This is the founder getting their eyes back. (That's what I'm building Foresight to do. In a box. Ready on day one. Not a 3-year management restructure.)

Most founders think they're losing deals on price. They're not. They're losing on legibility. A prospect can't tell the difference between three agencies pitching the same outcome in the same deck format with the same Slack-screenshot proof points. So they pick on price, because price is the only variable that's actually legible. The fix is not a better deck. The fix is making the work itself visible while it's happening, before the proposal goes out. Founders who figure this out stop competing on rate cards. Everyone else keeps discounting and calling it "market pressure.”

The thing you learn running a B partnership pipeline at Amazon: most deals that almost closed should not have. We had a system that could green-light a 9-figure agreement in 6 weeks if the spreadsheet looked clean. The spreadsheets lied beautifully. The deals that survived 3 years out were always the ones a single person inside the building hated, quietly, in a hallway. That objection lived nowhere on the deck. It lived in a Slack thread at 11pm, or a gut reaction nobody wrote down. AI is finally good enough to surface that in real time. Most orgs are still optimizing the deck.

Watching services founders kill their own AI rollouts by running them through committee. Six months in: three slide decks, two consultants, an "AI strategy framework," zero workflows actually changed. Meanwhile the founder who rewrote one process themselves over a weekend is already running on the new version, taking on bigger clients with the same headcount. AI doesn't reward consensus. It rewards the operator willing to be wrong fast and rewrite faster. Committee strategy in a 30-person firm is just middle management in a smaller suit.

Brilliant @blackmirror

Pro tip, you can cut off teslas because they brake automatically. So when you see a Tesla, cut it off.

FAANG has an IBM-in-1992 problem. Layer upon layer of middle management means layer upon layer of white lies. Eventually the execs can't tell which way is up. 5% growth and they call it a win. That's what easily disruptable looks like from the inside. Musk and Block are flattening. FAANG hasn't moved. Zuck is trying, I'll give him that, but Carmack went to Meta with all the reputation capital in the world and couldn't ship. The bureaucrats won. Apple and Amazon don't even have the founder anymore. This is exactly how Blockbuster, Sears, and Yellow Pages looked right before the internet dropped the trapdoor.

Cinco de mayo