Story time!
Rachel is a product designer at Cheddar, a budgeting app. She’s been working on a tricky problem: how to present spending alerts to users without making them feel judged or overwhelmed.
She can't tell whether there’s a real problem with the design or whether she’s just been too close to the work for too long. Her usual sounding board is on PTO, and she wants a few extra eyes on it before the design review.
The review goes fairly well. The team leaves a ton of comments on the work she presented. Now Rachel is back at her desk, staring at forty-odd comments scattered across the canvas. It's going to take a while to pull out the most important themes and figure out what to do next.
If you've ever been in a good design crit, you know that high-quality feedback comes from setting the stage before you show the work.
When you frame the problem, name what stage you're at, and specify what kind of feedback you’re looking for, you create useful boundaries. Without that context, feedback can come up about anything and everything.
The same is true when you work with an AI agent. An agent can help you get a sharper first pass before you bring work into the room. As a bonus, it can also help you make sense of everything that comes out of the review afterwards.
Get high-quality feedback from the agent
Just like setting the stage for a good crit, getting useful feedback from an AI agent depends almost entirely on the quality of the ask. Here are three ideas you can build into your prompts.
1. Give the agent a persona
Whose perspective should the agent take?
It could be a fresh perspective from a new user, a power user, or someone who’s had a stressful week. It could be a product leader who cares about monetization, or a head of growth who is focused on activation.
2. Scope your feedback
Which parts of the design do you need feedback on?
| Area | Questions to consider |
|---|---|
| Usability | Is the flow intuitive? What might confuse a user? |
| Accessibility | Does the contrast, tap targets, type size, and reading order meet accessibility guidelines? |
| Copy and microcopy | Is the tone right for the current moment? Is the content consistent with the product voice? |
3. Provide the right context
- What are you designing and who is it for?
- What are you trying to solve? What’s the goal of the work?
- What are you specifically unsure about?
- Are there any constraints the design has to respect?
Tip: If there’s a feedback prompt you find especially valuable, consider turning it into a skill.
Assemble a feedback prompt for this screen
Drag one or more of the prompt ingredients into the prompt zone to construct a helpful feedback prompt for this screen.
Turn feedback into action
Great feedback is a gift, especially if it offers a clear path forward.
After the agent surfaces issues, choose which ones to act on and ask it to make changes from the same thread. For example: "You flagged that the data labels have a contrast issue, please go ahead and fix those."
You don't have to translate critique into edits yourself; the agent already has the right context and can move straight to implementation.
You can also bring the agent in after a design review and ask it to help you make sense of the comments on the canvas. For example:
| Agent task | Example prompts |
|---|---|
| Summarize and cluster | "Check out the comments from my design review. Can you group these by theme and flag any that contradict each other?" |
| Prioritize | "Given that the goal of this screen is to drive two-week retention, which of these pieces of feedback are most important to address?" |
| Pressure-test a direction | "My PM wants to remove the chart entirely, but my manager thinks it's important. Here's the original brief; which direction is better supported by the goals?" |
Tip: You can even leave yourself comments on the canvas with a list of things you’d like the agent to improve!