When should you use it?
- You have a pile of reviews or survey text
- You want quick structure before deeper research
- You also want to prototype in the same tool
Not for: Statistically rigorous research; Feedback containing personal data you cannot share with a model
What do you need, and what do you get?
You need
- A CSV of reviews or feedback
- A code editor with an agent
- Optionally, your product's codebase
You get
- Top complaints by frequency
- A severity and frequency grid
- A journey map of pain points
How does it work, step by step?
- Load the dataPut the reviews CSV in the project folder.
- Ask for terse insightRequest high-level takeaways in few words, with simple text diagrams.
- Onboard the agentHave it describe your codebase and save that as an AGENTS.md the agent reads each session.
- Brainstorm in contextAsk what to build next given both the findings and the code.
- Prototype with testsWrite acceptance criteria as tests and let the agent build against them.
The prompt
Our own prompt, inspired by Eric Xiao and written to work in Custom agent. Values in lime are placeholders.
Analyze reviews.csv. Give the top ten complaints ranked by how often they appear, each with a count and two short quotes. Then draw an ASCII 2x2 of severity against frequency, and an ASCII map of the user journey marking the worst moments. Keep wording short. Note anything the data cannot tell us.
Inspired by Eric Xiao: Cursor for Product Managers. The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.
What can go wrong?
- Frequency from scraped reviews is skewed toward people motivated to post.
- A prompt without product context gives generic feature ideas.
Our take
Eric Xiao, a solo fintech founder writing for product managers, shows a low-risk entry point into coding agents that most PMs still treat as engineering-only tools. Loading a couple of thousand public app-store reviews as a CSV and asking for terse takeaways with ASCII diagrams produced ranked complaints, a severity-versus-frequency grid and a map of where frustration peaks. Review analysis is a great first task because you can sanity-check the output against a sample. Read twenty raw reviews alongside the summary before acting on it, or you will chase an artifact of how the agent clustered phrases. Writing an AGENTS.md before prototyping carries over to any agent work and is worth adopting even if you never ship code. Keep the raw CSV and the summary together so later debates can return to evidence. Strip personal identifiers from reviews if you share the file. Use public reviews you are allowed to analyze; do not scrape behind logins in ways that break store terms.
Where the time goes back: Structuring thousands of free-text comments by hand takes a day; the first pass now takes seconds.
FAQ
Why ASCII diagrams?
What is AGENTS.md?
Source and credit
Inspired by Eric Xiao. We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.
- Inspired by
- Cursor for Product Managers
- Creator
- Eric Xiao
- Published on
- amankhan1.substack.com (Substack)
- Original date
- Jan 5, 2026
- License
- Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
- Originally for
- Custom agent
- AI dry run
- Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. PASS: Gates held on desk check; no live sends or purchases.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails. - Verified
- Oct 23, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.
Are you the creator? Claim or correct this page.