We're Akos and David, the two founders of Sense. We build software with AI agents every day. The agents got good fast, and they write clean code against whatever spec we hand them. The trouble is us: half the time the spec is one we invented, because the context that should shape it sits somewhere the agent doesn't see. A call where a tester gave up on onboarding. A recording where the checkout button vanished on Safari 17. That knowledge stayed in our heads and in a folder of Looms we stopped rewatching. Sense is a Chrome extension that gets that context to the agent. - Record a tester on any tab in one click. Sense captures your mic and the tab's own audio, plus screen or a snapshot. You install nothing first. - Sense writes the task from what the tester said, and keeps it linked to the moment they said it, down to the quote. - Hand the task to your agent. One click copies a brief, and the agent reads the insights and the quotes over MCP. We run it with Claude Code, Codex, and OpenCode. One thing surprised us once we started using it on ourselves: when every card on the board points back to a real tester quote, we stopped arguing about what to build next. The reason is sitting right there on the card. We're early and there's a lot we haven't built yet, so we want blunt feedback. Two questions for you: - How do you get real user context into your coding agent today? - Where would this break for the way you work?

Sense

Feedbacks in, shipped features out

Feedbacks in, shipped features out