Monokit is on sale: the app screens most kits leave out · See Monokit → · Read the post →

← Studio

Built · 2026

Discovery without the sprint

A continuous scanner changes what discovery produces. Not depth on one area at a time, but breadth across everything, updated weekly.

Product opportunities don't wait for your next research sprint. NVSBL is a test of whether autonomous agents can keep scanning while you're building.

Product discovery gets treated like a project phase. You kick it off, run some research, converge on findings and move into execution. The signal is current for about a week. Then it starts to decay.

Most teams work that way. I think it's wrong.

Not because research doesn't work. Because it treats a continuous signal as a one-off input. Opportunities show up in forums, complaints and workarounds, and they don't wait for the fortnight when you happen to be doing discovery.

So I built NVSBL, to test whether an agent can keep looking while I'm building.

NVSBL — invisible agent platform for continuous product discovery

NVSBL — invisible agent platform for continuous product discovery

HN Scout reads the complaints

The first signal source is Hacker News, and specifically the complaints: Ask HN threads, "workaround for X", the posts that say "I built this because nothing else did Y". They're product opportunities nobody has packaged yet.

HN Scout runs weekly. It scans recent posts, runs them through an LLM and surfaces the ones that look like unmet needs. In the first working runs it scanned 284 posts and filed 30 labelled opportunity issues to GitHub, tagged by category.

Whether those 30 are worth building is a human question, which is why HN Scout doesn't try to answer it.

A person still decides

What the scanner produces isn't product. It's a list of candidates for someone to judge on commercial potential, competition and fit. Fully automated validation is a different and much harder problem. The agent's value is that it keeps looking, at scale, on a schedule.

Opportunities that pass review become the approved feed, served through a public API. Subscribers get them with an API key, and qualified findings can land as Linear issues in their own workspace.

Three repos, three evidence boundaries

NVSBL is three repos because each layer has a different bar for what can be shown:

  • nvsbl is the private engine. No UI: agents run and results go to GitHub Issues for review. Nothing leaves until it's approved.
  • nvsbl-web is the public surface: the approved feed, the request-access form and the API.
  • nvsbl-app is the authenticated layer for buyers: API keys, usage, billing, the Linear integration and pipelines.

Keeping them apart means I can show the site without showing the engine, and the engine can collect signal before the app is ready.

What the first runs taught me

HN Scout did what I expected. The scoring separated real pain from people complaining about life, and a run's issues took about 30 minutes to review.

What I didn't expect: the value isn't any single opportunity, it's the accumulation. After four or five runs the issue backlog starts to look like a map. Themes emerge, and the same need turns up from different directions.

A research sprint gives you depth on one area at a time. A scanner gives you breadth, updated weekly. It doesn't replace talking to users. It tells you where to aim.

The open question

NVSBL is live at nvsbl.dev. The scanner runs, the public feed exists, and the API is built but not yet selling.

The delivery loop is wired. What I'm testing now is whether a team of one can produce enough reviewed signal, fast enough, that product teams will pay for it.

NVSBL →