The Excellent Method

Every company already runs on the same handful of things — the work, the pipeline, the people, the money. Most software chops those into separate apps that each keep a partial, stale copy and never quite agree.

We think that's backwards. There is one business; it deserves one designed set of records that everything else is merely a view of. What follows is what we believe, and why the product is built the way it is.

Your work, your pipeline, your people, your money — in most stacks these live in five products that each hold a slice and none hold the whole. We keep them as one designed set of records on your own machine.

Not to lock them away — so that everything, and every agent, can see all of it at once. Ownership here isn't a pitch about vendor lock-in. It's a plain fact, said the way Obsidian says it: the records are on your disk, in one store, and you decide what leaves it.

A CRM is a way of looking at people and deals. An applicant tracker is a way of looking at people and jobs. Underneath, it's the same people.

So we don't ship five apps that each define 'a person' their own way — we ship one deliberately designed schema and let each surface be a lens on it. A candidate is a person. A lead is one or more people. A résumé is an attachment. You change the lens, never the data.

The industry's default is a chat window bolted to the corner — you ask, it answers, you paste the answer somewhere real. That's a demo, not a coworker.

Here an agent claims a task in the same column you would, moves it through the same states, and hands it back with a trace of what it did. You supervise one board, not two systems. The agent isn't a feature of the app; it's a participant in the work — and it uses the same tools you'd give a new hire.

Every change moves through four states — todo, in progress, review, done — and a different identity has to verify it before it lands. This isn't a guideline you're trusted to follow. It lives at the data layer.

Ask to approve your own work and the system doesn't frown — it refuses. The same rule binds a person and an agent equally, and that equality is the whole trick: an agent gets to touch real work precisely because it's held to the exact check you are.

Every agentic action is measured before anything is touched. A process is modeled as something you can evaluate, and a change only ships if it beats what it replaced — on evidence gathered offline, before it spends real money.

Even the optimizer is held to the gate: it can't promote its own work. And when a change misbehaves, a watchdog rolls the system back to the last known-good on its own. Improvement is a measurement, not an opinion.