The Questions Nobody Routed

Somebody had to read what came out of the machine, throw most of it away, and walk the few that counted over to a person who could still do something. Early enough that doing something was cheap. Gently enough that the question landed as a question and not an ambush.

That was my job. It had no name. It was in no scope I ever signed. It lived on no org chart, which meant that when budgets tightened, it was free, and things that are free are eventually treated as worthless, which is a sentence you can build a whole career on without noticing you're inside it.

Everyone could see the model. Nobody could see this. So the model got the credit, and I looked like overhead. Friendly overhead. Overhead you were glad to see coming. But overhead.

Here's the part I understated for years, mostly to myself.

Sorting live questions is the easy half. The hard half is knowing when a question has died.

A finding can be traced to the right sheet, dimensioned to the sixteenth, sourced like a dissertation — and be fiction. Somewhere a route moved. A scope shifted. A decision got made in a room the model never saw, by people who never told it. The question was real on Tuesday and dead by the following Tuesday, and nothing anywhere registers the change. The finding just sits there, confident, wrong, beautifully formatted.

You don't catch that with a better model. You catch it because you were carrying the decision history around in your head like a man carrying soup, and you felt the ground move.

Swap the instrument. Same story.

The estimate surfaces variances nobody reconciles. The schedule surfaces slipping float that nobody escalates until the float is a memory. The handover package arrives at the facilities office like a foreign language dictionary — here are a thousand answers to questions you have never asked and cannot pronounce. This industry has spent decades building better ways to see problems early, and almost no time deciding whose job it is to act on what gets seen.

Almost. There's one exception, and the industry's relationship with it tells you everything.

Production planning got this right. A constraint log with a name next to every line. A promise with a date on it. Someone runs the board, retires what's dead, walks what's alive over to the person who has to commit. Nobody calls that overhead. It's the whole reason it works.

And so naturally, half the industry avoids it like a summons. The other half performs it. The meeting occurs. The board is populated. The stickies go up, and everyone stands, because standing means lean. Nothing binds. Ask about it afterward and you get the same answer everywhere, delivered kindly, by good people: we basically already do this, our own way.They don't. But the sentence isn't a lie exactly. It's a kindness they're doing themselves, because saying yes out loud would mean admitting what everything else has been.

We'd rather tell the story about the superintendent who held it all together because he was that good. I wrote about him already. He was that good. That's the tender part. He was that good, and he should never have had to be.

If the sorting and the walking-over is the asset, then the tooling isn't. Neither is the department built around the tooling.

I would have fought that sentence for most of my career, politely, in meetings, with slides. It threatened the seat I was sitting in. When your paycheck rides on production hours flowing through your team, defending the craft and defending your position feel exactly alike from the inside. There is no inner instrument that tells you which one you're doing. I checked. There isn't.

I got clear on it after the seat was gone, which is the only time anyone gets clear on anything.

Some shops never had this argument at all. Their model feeds a machine — a spool drawing runs a cutting table, and the table doesn't attend meetings. Nothing argues with the thing that feeds the shop. Where the output feeds a meeting instead, the same capability spent twenty years being advisory. A meeting can be declined. A meeting can be rescheduled forever. A meeting can nod.

Your model surfaces questions. So does your schedule. So does your estimate. AI is about to make all of them louder, faster, and more confident.

Who decides which questions matter? Who retires the ones that quietly died? Who makes sure the right person is expecting the rest — expecting them, not bracing for them?

Most programs can't put a name to that. Almost all of them could. It has to be somebody's job before it can be somebody's name.

Until then the instruments keep getting better. And somewhere on every job there's a person doing this work anyway, unnamed, unfunded, carrying the soup — and the whole thing quietly rests on whether they stay.