Product feedback for Bolt

This page is the playbook for getting visual feedback into a Bolt session without slowing the iteration loop. For the same pattern in other tools, see the full feedback for AI coding agents hub.

The problem with feedback in Bolt workflows

Bolt rewards short cycles. You build, preview, ask for changes, rebuild. The cost of feedback that takes ten minutes to file is not just the ten minutes, it is the broken rhythm. By the time you've written up six findings in proper prose, the build is stale and you've forgotten which screen the third one was on.

The other failure mode is the opposite: you keep iterating without structured feedback at all, just typing one-liners in the chat. Bolt fixes the thing you mentioned and silently undoes a fix from the previous cycle because there is no record of what was working before.

What you want is something fast enough to keep up with Bolt's iteration speed and structured enough to be a record across rebuilds.

The CobaltCapture workflow with Bolt

Bolt preview open in one tab, cobaltcapture.com in another. Hit Capture screen, pick the Bolt window, crop to the issue, dictate. "The empty state for the projects list shows the wrong icon and the copy uses sentence case where the rest of the app uses title case." Move on. Capture the next finding. Three to five findings in two minutes.

Hit Publish. Drop the URL into Bolt's chat:

Review for this iteration: https://cobaltcapture.com/r/<slug>

Apply the three fixes. After each, confirm the change and which file
you touched. Don't regress the empty-state logic from the previous
iteration.

The published review also serves as a record. The next time something regresses, you have a frozen URL of how it looked when it was working.

Why this works for Bolt

Bolt's chat input takes URLs and follows them. The iteration loop stays in the chat, no context-switching to a ticketing tool, no copy-paste theatre. And because the review URL is a permanent artifact, you accumulate a thin record of decisions without anyone consciously documenting them. Each iteration leaves a footprint.

Dictation matters here because Bolt's prompts are short by design. "Fix the empty state" gets you a different patch than "fix the empty state, it should match the title-case convention from elsewhere in the app, and the icon should be the same one we use for the dashboard empty state." Saying it out loud is fast. Typing the same thing breaks the iteration rhythm.

Alternatives and tradeoffs

You could keep typing one-liners into the Bolt chat. Fast for one change, lossy for five, and zero record.

You could pause Bolt iteration and write proper tickets in Linear. Tickets are great when work is going to take days. Bolt iterations take minutes. The overhead of a ticket per iteration kills the loop.

You could record a Scribe walkthrough of the issues. Scribe is built for documenting human workflows, not for feeding feedback to an AI agent, the scribe alternative angle. Its HTML output is trapped in Scribe's app and is not pasteable into Bolt.

CobaltCapture is the artifact that matches Bolt's pace and produces a record at the same time.