I've met a lot of customers in person and it's crazy to me how most founders have no clue what their customers are asking for. You can't fix your product, onboarding or roadmap until you know what's really broken. I've automated most of this at Polsia. Here's my setup: 1. Report during the conversation Agent log every issue while talking to customers: > Is it a real bug? > Is it a feature request? > Is the customer just confused? It's a direct part of support so everything is logged. 2. Read every ticket A PM agent pulls all tickets from the day and ranks how critical it is. It's usually not a bug at all. 3. Build everything with an agent Top tickets go straight to an engineering agents that builds the fix. So we don't just know what customers want. We ship the fix as well. 4. Check what it built against what might break We QA each fix, then ask Codex to double check it's safe to ship. The build is easy so focus on verifying it's right. 5. Turn it into scope Agents only push to production on their own when every rule is met: > Severity: lower-category bug > Payments: fix doesn't touch them > Onboarding: fix doesn't touch it either > Second opinion: Codex agrees it's safe > Everything else: comes through me first Then we just widen the scope as models get better. This runs our bug fixes, features, releases and roadmap.