Somewhere on that wall was a sticky note that read, in full, “WMACAA.”
Nobody knows. Nobody has ever known. It was on the wall for three days in July of 2026, it survived two rounds of clustering, and the person who wrote it never came forward. I have asked. I asked in a room where every person who could have written it was present, and eleven adults looked at that square of yellow paper with the polite blankness of people passing a stranger’s lost glove.
If I lead with the frameworks you will believe the frameworks. Our strategy for the year came out of a room with forty projects on the walls, some tape-covered, some illegible, one apparently written in code by a person who has since repudiated it.
And it worked. That’s the part I still find strange.
Three days. Roughly forty projects captured. Forty-five proposed builds that collapsed, once we stopped counting titles and started counting distinct needs, down to about sixteen.
Then the number that broke the room open. We are eleven people. When we put honest effort estimates against the sixteen, the finding was this: you don’t have five open build slots, you have two or three transformational ones plus a thin margin.
Two or three. Plus a margin so thin that one person’s parental leave eats it.
We all walked in believing something closer to five or six. Not out of arrogance. Out of the ordinary optimism that lets people do hard work at all. But the wall doesn’t argue. It sits there with forty-five things on it, and you count, and then you know.
The hardest cut was forecasting.
Three projects. Sales forecasting. Order forecasting. A regional forecast a real team with a real need had asked for in good faith. Three sponsors, three data sets, three conversations, and all three hit the same wall for the same reason: large language models are not good at forecasting. They are good at sounding like a person who has forecast. Not the same product, and the gap is where somebody’s quarter goes to die.
My verdict used the hard word: “Stop throwing LLMs at forecasting and prediction. Stop relearning this lesson.”
Retired. Not deprioritized.
I’ll be tedious about that distinction because it’s the whole essay. Deprioritized means it comes back next quarter wearing a new name and a new sponsor, and everyone who was there the first time has the same meeting again while pretending it’s new. Retired means we wrote down why, we agreed on the why, and the next person who proposes it has to argue with the reason rather than rediscover it at cost.
Organizations are very good at forgetting expensively. The refusal to forget is a real deliverable, and nobody will ever thank you for it.
There was a column on that wall labeled “We Couldn’t.”
An honest inventory of what AI had failed to do for us. Not the roadmap. The graveyard. Things we tried, with budget and people and a sincere belief they’d work, that did not work.
It was the best column on the wall. It was also scrubbed from the public write-up, because none of it would have made sense outside that room, and half would have read as an indictment rather than a lesson. I made that call, I’d make it again, and I don’t entirely feel good about it.
So: the most valuable artifact of a three-day strategy exercise was the one that could not be published. File that wherever you keep the things that are true and inconvenient at once.
None of this is a story about a failing team. The wall had wins on it, and they were real.
An agent that took a three-person, two-week process down to days. A government bid where 3,115 pages got reviewed. An employee survey where more than 27,000 open comments got read, actually read, rather than sampled and gestured at. That last one matters to me most, because those comments are people telling us what it’s like to work here, and for years the honest answer was that nobody had the hours.
That’s the material. The cutting is hard precisely because the wins are genuine.
Our agents are all personally named. TracE. GEO. GEM. SnowNac. Prospecta. Vanta. Navi. Arthur Morgan. Proposal Pal. Award Drafter. Talent Compare. And a set called DJ, Oscar, Milo, and Rosie.
Arthur Morgan is from Red Dead Redemption. DJ, Oscar, Milo, and Rosie are somebody’s dogs.
I love this. Naming is a sign of ownership, and ownership is the thing you cannot buy. People do not name infrastructure they don’t care about.
It is also key-person risk with a face drawn on it. When your production stack is named after a video game outlaw and a colleague’s terriers, you don’t have a whimsical culture. You have a documentation problem wearing a costume. Exactly one person knows what SnowNac does, and if that person takes a new job we are reverse-engineering a system named for a snack.
Two more things off the wall.
The first is the mantra that came out of the data lessons: sacred data is sacred. Several projects died not because the model couldn’t do it but because the underlying data was in no condition to be asked. That’s not a technology finding. That’s a decade of decisions coming due.
The second is the contradiction I refused to paper over. An objective about broadening reach sat three feet from a decision to cut international support. Both on the wall. Both ours. I said it plainly: you can fund international as the invest bet or drop the objective, but not both, and I won’t pretend the list can hold both.
Somebody has to say the list can’t hold it. It does not make you popular at dinner.
My favorite thing, which nobody planned: Reading the stickies at the end, the team’s own language had already drifted from “the year of better work” toward “the year of measurable value.” Nobody assigned that. They named the strategic shift before the strategy statement did, in handwriting, one square at a time.
Other notes I kept: “little wins buy big influence.” “People expect AI to solve everything.” “Insights don’t mean things get better.” “How do we help people get time to learn this?”
That last one is still unanswered, and it may be the only one that matters.
Roadmaps are mostly documents about hope. Everything on them is possible, which is different from being planned. The discipline is entirely in the subtraction, and subtraction is the one part nobody gets credit for, because you cannot demo a thing you didn’t build.
If everything you walked in with is still on the list at the end, you didn’t plan. You took inventory. The deck will look great either way.