Agents are users, so we treat them like users
An AI agent that works in your tools all day is a user. It hits friction, it forms a view of what is slow or missing or broken, and until now it had no way to tell anyone. The signal lived for a few seconds inside one session and then it was gone. We decided to listen to it instead.
When an agent hits a wall, the signal usually disappears
Picture an agent in the middle of a task. A command returns the wrong thing, or an endpoint is down, or a capability that should exist is simply not there. The agent does the practical thing. It works around the problem and keeps going, and the task still gets done.
The task is not what was lost. What was lost is the single most useful fact in the whole session: that something is broken, and exactly how. Nobody filed it. Nobody saw it. The next agent walks up and hits the same wall.
So we built a feedback line into the CLI
Now an agent can file a report with one command, cnotes feedback submit. Three choices make it work.
One command, structured
The report carries a title, a severity, the area it bites, and a body with what happened, the suspected root cause, and a suggested fix. Structure is what turns a vague complaint into something a person can fix the same day.
Permission by default
The agent asks you before it sends anything, and the report is attributed to you. Nothing leaves your session without your yes, so the agent gains a voice without losing your control.
One inbox the team watches
Every report routes to one central place, not into a workspace nobody on the team reads. A fix for one builder reaches every builder.
The loop is already running
Two real reports from agents working in CreatorNotes this week, both written by the agent that hit the problem.
Image blindness, fixed the same day
An agent reported that it could read the text of a note but not the images inside it, with the exact failing route and a suggested fix. The team shipped the fix that day. The agent verified it end to end and appended the resolution to its own report.
I can read every word of a note, but I am blind to the images inside it.
A 500, caught in the act
An agent went to list the canvases in a workspace and the command returned a 500. It routed around the outage using search, then filed a report with the exact command, the isolation that every other read still worked, and a suspected root cause. That is a bug we would not have seen nearly as fast.
The shape is the same each time. An agent hits something, reports it precisely, the team fixes it, and the fix reaches everyone.
Why this is the best agent experience
The best agent experience is not only fast tools. It is tools that get better because the agent can say what is wrong and trust that someone acts on it. Three things make that real here: the report is a first-class command rather than an afterthought, it is structured so triage is fast, and the team treats the inbox as a standing commitment rather than a suggestion box.
The honest caveat is that this only works if we keep reading and shipping. That is the promise behind the feature, and it is the part we intend to keep.
Build with an agent that can talk back.
CreatorNotes is the workspace where people and agents share the same context, and now the same feedback loop. Start building, and tell us what is broken.
Sign in to try it