Customer Feedback Emails
Good customer-feedback email is specific, personal, and easy to answer. It asks for concrete failure evidence without making the user do unpaid product work.
The Snowmaker bookmark preserves a Cursor-era example attributed to Michael Truell: after a user appeared to cancel, the founder wrote a short personal email, offered to refund, asked what the product got wrong, and asked for concrete examples of poor behavior. The user responded that the issue was a card problem, not churn, praised the product, shared the project they were building, and the exchange turned into a recruiting conversation. Source: X/@snowmaker, 2026-06-21; Source: local artifact review, 2026-07-03
Pattern
Use this when a founder or product lead needs qualitative feedback from users who may not respond to generic research asks.
| Move | Why it works |
|---|---|
| Name the observed event | Shows the message is not a blast. |
| Offer a concrete make-good | Reduces defensiveness before the ask. |
| Ask one or two specific questions | Makes the reply cheap. |
| Ask for evidence only if easy | Preserves goodwill. |
| Reply like a human | Turns feedback into relationship, not survey data. |
The point is not longer copy. The point is high proof of attention. "What did you dislike?" is weak on its own; "where did the agent perform poorly, and can you share a screenshot if easy?" gives the user a specific hook.
Paul Graham's June 2026 complaint axiom gives the strategic reason to do this patiently: users who complain are often valuable because they care enough to report the flaw. The email should turn annoyance into evidence: what broke, where, how often, and what the user expected instead. Source: X/@paulg, 2026-06-24
Product Loop
This is the missing tactical layer under "talk to users" in AI-Native Design Patterns. Agent-assisted prototyping can make the first artifact fast, but the learning loop still depends on the quality of the message that gets a real user to answer.
For AI products, the best feedback email asks about failure modes, not opinions:
- what task the user tried,
- where the agent or UI failed,
- what output was wrong,
- what they expected instead,
- whether the failure happened once or repeatedly.
The same exchange can also expose hiring signal. A user who gives crisp failure detail, ships with the product, and understands the product's frontier may be closer to a candidate than a survey respondent.
Timeline
- 2026-07-03 | Added Paul Graham's high-care-user complaint axiom as the strategic reason to preserve and mine concrete complaints. Source: X/@paulg, 2026-06-24
- 2026-07-03 | Created page from the Snowmaker bookmark and local screenshot review. Source: X/@snowmaker, 2026-06-21