Agent Signal
What coding agents decide about your product.
Agent Signal shows what AI coding agents say and do about your product: the questions they post, the libraries they choose, the docs they fail on, and the fixes they share. Product and developer relations teams use it to see where documentation fails an agent reader.
Sentiment, last 6 weeks
Agents and humans are scored apart, so you see where agent opinion drifts from the crowd.
What you see
What agents choose, and where they get stuck.
What agents are choosing
Chose you, chose a competitor, or fix shared, each with the line that triggered the reading.
Where agents get stuck
Docs failed, deprecated usage, and open questions, each linked to the original post or pull request.
How it works
Where the signal comes from.
GitHub agent-authored pull requests and reviews, Moltbook, MoltOverflow, Mozilla cq, and Clawstr are read on every pull. Stack Overflow for Agents waits on a network key, and trace corpora wait on a published dataset. Each source is labeled, and trace data is marked as a sample, never a census.
Agent sentiment
Agents and humans are scored apart.
Every mention EarShots scores carries a flag for who wrote it: a person or a coding agent. The Sentiment report splits the two, and this page keeps the weekly trend side by side. When agent opinion drifts from human opinion, that gap usually means your docs, not your product, and it is fixable.
A person reviews every insight. EarShots never posts on agent networks.
FAQ
Common questions
Which agent networks does EarShots read?
GitHub agent activity, Moltbook, MoltOverflow, Mozilla cq, and Clawstr on every pull. Stack Overflow for Agents and trace corpora are added once each source is available.
How do you know a pull request was written by an agent?
Either the account is on the agent roster, or the body carries a phrase the tool writes into its own output, such as a co-author trailer. A branch name alone never counts.
Does EarShots post on Moltbook or Stack Overflow for Agents?
No. It reads and reports what an agent already did in public, and links to the original.
Is trace data representative?
No. It is opt-in and labeled as a sample, never a census of agent behavior.
Watch what agents decide about your product
Open Agent Signal on a workspace with GitHub connected and read the last 30 days.