Your documentation

Nobody wrote the automation.
It was already there.

Almost every attempt to let an AI agent operate business software, computer use, as it is usually called, works the same way: photograph the screen, work out from the pixels what is on it, click something, photograph it again. FlowShare does none of that. It never looks at your screen and it never infers a button from how it looks. It works from what the application itself reports about its own controls.

That is possible because of something we have been building for years, in plain sight. Every time someone documents a process with FlowShare, they are also capturing everything an automation needs: every click, every entry, and the identity of every control involved. It costs them nothing extra, because they were writing the guide anyway. The automation data is a by‑product of work your team already does.

So what never changes stays fixed, and the agent decides only what genuinely varies: the one value that is different this time, which branch the live screen calls for, which recorded procedures to string together for the job in front of it. Because controls are identified rather than located, a redesigned screen or a new software version does not quietly break the run. And when a target is genuinely ambiguous, FlowShare stops and says why instead of clicking and hoping.

Your documentation and its execution stay on the machine. Recording is local, replay is local, and your screen is never transmitted anywhere for a model to interpret. The agent is told what to do next; it is not shown a picture of your desktop, your other windows, or whatever happened to be open beside them.

And because there are no images in the loop, the agent’s job is a small one. No vision model, no screenshot per step, no long chain of image tokens, just a short exchange of text about what to do next. Token consumption stays low, and so does the bar for the model driving it, which is what makes running this at the scale of a real process affordable rather than a demo.

“A straight word on where this stands. FlowShare for documentation is a shipping product that thousands of teams already use, and none of this changes it. The execution layer is newer: how completely it covers an application depends on how that application exposes its interface, and a long process will still sometimes stop and ask for a human. Which is why we are looking for a small number of partners with a real process to test it against, rather than selling a finished product”.