Guide

How to write website feedback that an AI coding agent can act on

Feedback an AI coding agent can act on states what to change, where it is and what “done” looks like, and ideally carries the page URL, element and screen size. Vague comments cost extra rounds; a pinned, specific comment is usually fixed in one pass.

Updated · 6 min read · by the Symu 2 team

The anatomy of an actionable comment

An AI coding agent can do a lot with a well-formed instruction and very little with a vague one. Good website feedback has four parts:

  1. Where. The page and the element. If you use a tool that pins comments to elements, this part is automatic.
  2. What. The change you want, in plain words.
  3. Done when. How you will recognise success.
  4. Constraints. What must not change, such as brand colors or layout on other screens.

Vague versus actionable: examples

VagueActionable
“The header looks off.”“On mobile the logo overlaps the menu button. Keep the logo on the left and give the menu button 16px of space.”
“Change the button.”“Make the Sign up button green (same green as the badge) and change the text to ‘Start free’.”
“Too much text.”“Shorten this paragraph to two sentences and keep the pricing figure.”
“Doesn’t work on my phone.”“On a 390px wide screen the pricing cards overflow to the right. Stack them in one column.”

A short template

Element:  what you pinned (the tool records this)
Change:   one specific change
Outcome:  what it should look like or do afterwards
Keep:     what must stay the same

If you pin the comment with Symu 2, you only need to write the “Change”, “Outcome” and “Keep” lines. The element, page URL and breakpoint are stored with the comment and passed to the agent.

Batching comments for an agent

Agents work best on a queue of small tasks. Collect all comments for a review round, export the open ones as a markdown task list (or let the agent read them over MCP), and ask for one fix per comment with a short reply. This also gives you a clean record of what changed and why.

Verify, don’t assume

Ask the agent to describe what it changed, reload the page at the same breakpoint and compare with the outcome you wrote. Resolve the comment only when you are satisfied. If the agent misunderstood, reopen the comment and add the missing constraint instead of starting a new thread.

Next: the full workflow with Claude Code or the vibe coding feedback tool overview.

FAQ

Frequently asked questions

More questions? Write to contact@symu.co.

What makes feedback actionable for an AI agent?

It names the page, the element and the desired outcome, and says on which screen size the problem appears. When the feedback tool stores the selector, HTML and breakpoint with the comment, the agent gets most of this automatically.

Should I write feedback as a prompt?

Write it as a plain request in your own words: one change per comment, a clear outcome and any constraint. A feedback tool that wraps your words with the element context, such as Symu 2’s “copy as AI prompt”, produces the final prompt for you.

How do I verify that the agent fixed it?

Ask the agent to reply on the comment with a summary of what changed, then reopen the page at the same breakpoint and resolve the comment yourself once it looks right.

See how Symu 2 works on your own website.

Paste a URL, invite a client and let your AI read the feedback. 14 days free, then $19/month.