Pathmode

Over the past year I've watched teams (including mine) get dramatically faster at building. Agents write the code, ship the PR, fix the bug. Work that used to take a sprint takes an afternoon.

And yet — a lot of what gets built still isn't the right thing.

That's the part AI didn't fix. It made building cheap. It didn't make us any better at deciding what's worth building. When code was the bottleneck, sloppy judgment was hidden by slow delivery. Now that delivery is instant, bad judgment ships at full speed.

Most tools don't help here. Jira and Linear track when something gets done. Specs get written once and rot into stale handoffs. And an AI agent will cheerfully build exactly what you point it at — including the wrong thing, faster than ever.

So I built Pathmode. The idea: take the user evidence you already have — friction, quotes, support tickets, metrics — and turn it into intent: specs that carry the why, not just the what. Those specs become the source of truth your agents build from. I've started calling the discipline Intent Engineering — intent as judgment under evidence, not a contract you compile.

It launched about a week ago and I'm still early. I'd love feedback from anyone here building products with AI.

One question I keep chewing on, and I'm curious how you think about it:

When an agent can build almost anything you ask, how do you decide what's actually worth building?