Picture a big old house mid-renovation. Twelve rooms, wiring nobody trusts, a kitchen that's been "almost done" since spring. You've hired six people: a plumber, two electricians, a drywaller, a painter, and your cousin who's "handy."
Here's the problem. None of them know what the others did yesterday.
The painter shows up Tuesday and paints over a wall the electrician hadn't finished wiring yet. Both electricians, working different days, rewire the same outlet because neither one thought to check. Your cousin spends an entire Saturday hanging drywall in the one room that was already done, because nobody told him otherwise. By October you have no idea which rooms are actually finished and which just look finished until someone opens a wall.
What that house needs isn't more workers. It's a foreman who has physically walked into every single room, written down exactly what's done and what isn't, and hands out jobs so two people never grab the same one.
Lacunari is that foreman. Except instead of a house, it's your project (a codebase, a pile of documents, whatever you're building), and instead of people with tool belts, the workers are usually AI assistants. Sometimes both.
It starts by actually looking
Most tools that are supposed to help you manage a big messy project have never actually looked inside it. They just keep a list of file names, the same way a moving company keeps a list of boxes without knowing what's in them.
Lacunari opens every box. It reads the whole project, whatever kind it is, and builds an honest map: what's here, and what's quietly missing. Not a guess. An actual walkthrough.
Then it writes the punch list
Once it knows what's really in the house, it can tell you what isn't finished. Not "here's everything," which is what a search bar gives you. The other thing: which rooms have no working outlets, which walls were never painted, which pipe nobody ever hooked up.
That punch list becomes the actual to-do list. Not one somebody wrote from memory three weeks ago and never updated. One built from what's really standing in front of you, right now.
Nobody grabs the same job twice
Here's the part that matters most. When a worker, human or AI, is ready for their next job, they ask the foreman for one. The foreman hands out exactly one job to exactly one worker, and everybody else immediately sees it's taken. Not "taken, probably, unless somebody forgot to update the spreadsheet." Taken, for real, checked at the moment it's asked for.
If a worker wanders off and never comes back, their job doesn't just sit there forever marked "in progress" while nothing happens. It goes back on the list so somebody else can pick it up.
And when a job is genuinely finished, it comes off the list by itself. Nobody has to remember to cross it out.
Big vague goals become an actual checklist
You can also just tell it the goal instead of the tasks. Something like "get this place move-in ready." Lacunari breaks that down into real, specific jobs based on what it actually saw when it walked through, room by room, and hands those out the same way. Nobody's guessing what "move-in ready" means at three in the morning.
Why this is genuinely exciting
If you've played with one of these AI assistants, you already know they're remarkable at doing a task. Ask one to write something, fix something, or research something, and it'll often surprise you.
What almost nobody has solved is getting several of them to work the same project at once without wrecking each other's progress. That's not a small gap. It's the difference between one very capable assistant and an actual team.
Lacunari runs the clipboard. It doesn't swing a hammer, and it's honest about that. But once you've got a crew, even a small one, having someone who's actually walked every room and never hands out the same job twice turns out to be the whole ballgame.
Next: the toe-stepping you're already dealing with → · The technical version, if you want it