AI codes 100x faster; the company still runs two-week sprintsEverything Is Changing, and Nothing Is Changing

AI capabilities are racing ahead for individuals, but inside companies they stall against two walls: one broken link spoils everything, and the organisation itself is a mess nobody fully understands.

A business-school professor and economic sociologist says his reaction to how fast AI is landing is not 'slow' but 'surprisingly fast'.

Superhuman alone, stuck together

This phase is single-player: one person can vibe-code an entire game overnight and write papers faster than ever.

Step into a system where many people work together and the gaps show. A coder is a hundred times more productive but still inside a two-week sprint that needs human input every fortnight. Huge token spend, no visible impact at company level.

Rocket parts and garbage cans

When Challenger lifted off in 1986, all sixteen thousand parts worked that day. One O-ring failed and the whole thing was lost.

Work is the same. If AI cannot do one or two parts of a job, you still need a human to step in — it does not add up and cancel out. AI writes code now, and engineers still have to manage it.

And what is a company? Organisational theory's least flattering model: a bundle of problems and solutions piled in the same garbage can, looking for each other. The plan is a small part; plenty of routines are dead ends nobody remembers creating.

So 'add AI to this task' reads cleanly on paper and falls apart once it hits the real garbage can.

The agents start acting

Personal agents like Meta's Muse dig through your email to find problems themselves. Stuck on an unexplained American Airlines subscription fee, the professor's agent joined the customer-service chat, cancelled it and filled in a follow-up form.

That friction is the business model: banks earn on low-interest accounts, and hard-to-reach support keeps refunds down. Agents tear straight through it. Amazon has already banned Muse.