We’re agile. Why doesn’t it feel like it? We’re agile. Why doesn’t it feel like it?

Stand-ups every morning. Sprint reviews every two weeks. A board full of tickets that move from left to right on schedule. And yet the thing that’s supposed to ship — the AI feature the business is waiting on — keeps slipping. The ceremonies are full. The software is empty.

• Velocity is up, output is down.
• Process is loud, decisions are quiet.
• Everyone is busy, nobody can say when it’s done.

A notebook convinced the room. A demo went well in a Teams call. Somewhere between that moment and the first real user, the answer got slower, stranger, more expensive — and nobody can point to the file where the regression lives. The gap between “it works on my laptop” and “it works for ten thousand users” is where most AI initiatives quietly die.

• The prompt that passed review cannot pass load.
• No one owns the evals. So nothing is really “better.”
• The cost per answer is a rumour, not a number.

Burn-down charts descend politely. Satisfaction scores float near the top. Requirements coverage is a comforting 92%. Meanwhile, the actual users are waiting for something that hasn’t moved in three months, and the steering committee is about to ask a very uncomfortable question at the next review. Metrics are measuring the ritual, not the outcome.

• Good numbers, bad product.
• Reports are written for the reporter.
• Nobody looks at the real user journey, because it would hurt.

A big-brand consultancy came in, ran a slick kickoff, left behind a seventy-slide deck and a pilot that wowed the execs. Then the junior consultants rotated off, the champion changed role, and the code became the team’s problem to carry. The polish didn’t survive contact with the backlog. Now the internal engineers are fixing things they didn’t design and can’t quite explain.

• The presentation outlasted the product.
• Knowledge left when the badges did.
• The deck is still beautiful. The system is not.

The staffing firm sent four people in week one, two more in week three, all billable by the hour, all waiting for someone to tell them what the right answer looks like. The hours tick. The tickets close. The architecture hasn’t been decided yet. An AI system built without a few hard, opinionated calls at the start is very expensive scaffolding around a very expensive hole.

• Headcount solved the wrong problem.
• No one on the contract is paid to say “stop.”
• The real decision keeps getting scheduled for next week.

They’ve read the papers. They’ve prototyped on weekends. They could probably get this over the line — if they were allowed to stop doing four other things at once. What they don’t need is another motivational kickoff. What they need is a senior engineer alongside them who has shipped this kind of system before, who can take the hard calls off the lead’s desk, and who leaves before the dependency becomes a habit.

• Capable people, out of runway.
• Context-switching has replaced engineering.
• They don’t need cheerleaders. They need a peer.