Start here
Why does Discover Agents stop at the test and not build my agent?
Why it works this way
The expensive mistake is building the wrong thing, not building slowly, so Discover Agents puts its work into the deciding and leaves the building to you.
You can build with the tool you already use. What is hard to do alone is finding out whether the thing is worth building: reading the field, writing honest tests, talking to people, and not fooling yourself with the answers. So the product does that part and stops where the building starts. The choice of building tool stays yours.
What we gave up: a button that takes you from your board to a finished agent. A founder whose test passed has to do the building somewhere else.
What it is
Discover Agents takes you from a first idea to a decided test. It writes the tests, helps you work out who to ask and reads what they say. It does not build your agent. When a test needs something built, such as a rough screen to put in front of people, you build it in the tool you already use.
How to use it
- Do the checking here. Pick a shape, write the tests, run them with real people and call the result.
- If a test needs something built, build only that, in your own tool. When you are ready to build, Discover Agents writes the build prompt from your own board, so you do not have to brief a coding tool from scratch.
- Bring back what happened and write it down.
- Decide when the evidence is enough. A pass means the idea has earned more work, not that the agent works. Discover Agents then writes the next tests from what the first ones found.
- When you choose to build the real thing, the building is yours.
What it will never do
- Build or run anything inside Discover Agents. It hands you text, and the building happens in your tool.
- Run your test for you. A test is a conversation or a measurement you carry out.
- Tell you the agent will work.
- Choose your building tool.
If something looks off
- You expected a finished agent. Nothing is built inside Discover Agents. What you get is the Blueprint, the tests and what real people said.
- A test says to build something first. Build the least it needs: a rough screen to put in front of people, or a small rig that runs real inputs through the risky step. Not the whole agent.
- You have no coding tool yet. Many tests need nothing built. An interview, an expert interview, an existing-data check and a cost check are conversations and arithmetic. See What kinds of test are there?.
- You want to build the whole thing and skip the tests. Nothing here stops you, and the Blueprint still says what might trip it up.