Right now every product team is wrestling with the same question: How do we actually build AI products that work? The advice is often too abstract or too technical. What’s missing is the messy middle: the product decisions, team dynamics, and quality trade-offs that determine whether an AI feature delights users or gets abandoned after one try.
Over the past few years, Productboard has re-architected its product experience to be AI-first. Work on Pulse and Spark taught us crucial lessons about quality and user trust—and about rethinking the product management workflow, not just bolting AI features onto an existing platform.
What this essay explores
- Validate both customer pain and whether AI can meet the quality customers expect.
- Decide when AI should automate a task and when it should help users think better.
- Match the interface and workflow to how users actually work, not default chat.
- Balance early-adopter learning with the polished, predictable outputs later customers expect.
- Set quality thresholds and systematically improve AI accuracy through customer-rooted test scenarios.
- Evolve from a small core team to specialized teams without losing shared context or velocity.
AI quality needs explicit thresholds before a feature reaches broad release.
Key takeaway
A detailed look at how one team navigated the hard calls: figuring out where AI actually fits, shipping ahead of customers, and building quality through rigorous evaluation & iteration.
