AI agents write the code. Then they go blind.
The error on mobile Safari. The page that got slow. The users who gave up.
Someone has to notice and paste it into a chat window.
I'm building JAMP so the agent can just look
🧵👇
So I'm building the other half. Errors, performance, uptime and real user behaviour, readable by your agent through MCP.
We taught machines to write code. We didn't give them eyes.
3/ So I'm building the other half. Errors, performance, uptime and real user behaviour, readable by your agent through MCP.
We taught machines to write code. We didn't give them eyes.
2/ The writing half got very good very fast.
The half where you find out what actually happened didn't move. It's still a human reading a dashboard, deciding what to tell the agent next.
The best way to test an analytics product is apparently to spend money on Google Ads and then use your own analytics product to figure out where the money went.
JAMP is getting a lot better very quickly.