The Intent Layer
For all of history, execution was scarce. Ideas were cheap, doing was expensive, and so every productivity tool ever built manages doing. AI broke that arrangement, and almost no software has noticed.
When something becomes abundant, value migrates to whatever stays scarce. Execution is becoming abundant: code gets written, copy gets drafted, research gets run, and the marginal cost of another task completed keeps falling. What stays scarce sits entirely on the human side: knowing what you want, saying it clearly, choosing what matters among a thousand things that are now all possible, and staying oriented while ten delegated threads move at once.
The felt experience of this shift is strange and widely shared: more gets done than ever, and people feel less on top of their work than ever. That is not a personal failing. The bottleneck moved. It used to sit between the plan and the work; now it sits between the person and their own intentions. Lists, boards, and tickets were built for a world where you were the execution engine, so they keep handing you execution-sized pieces: forty tickets, twelve dashboards, one overwhelmed human.
Someone will own the layer above the tools
Execution tools are becoming commodities that agents drive. The interesting question is what sits above them: the layer where a person holds the shape of what they are trying to do, decides what happens next, and delegates the method to whoever executes best, human or machine. Call it the intent layer. It is where goals live, where trade-offs get made, where "not now" is a decision rather than a guilt pile, and where progress from every executor rolls back up into one picture.
Chat cannot be this layer. Chat is amnesiac: every conversation re-explains your life from scratch, and yesterday's brilliant plan is buried forty scrolls up. Documents cannot be it either: they hold words, not structure, and they do not know what is done. The intent layer needs to be durable, spatial, and alive: durable so it is still true on Thursday, spatial so a human can actually hold the whole of it, alive so the work moving underneath updates it without anyone writing a status report.
What we are building
Orient is our attempt at the intent layer. Plans live as nodes on a map, the map keeps score as work moves, and everything an operator can do on the canvas, an agent can do through the API, under the same permission model. Your agents stop needing another chat window; they need your map: it is their brief and their report. You state the intent and hold the shape; the detail is machine territory now.
The wager underneath the product is simple to state: in an era where anything can be executed, the person who knows what they want is the rarest and most leveraged person in the room. Software should be making more of those people, and that is a different job than making anyone faster at tasks.