How OpenAI Product Managers Feed Data to Research
Tara Seshan explains how OpenAI PMs use evals, prompts, and sample readings to train user needs directly into base models.
Where AI products go next: voice, agents, and self-driving software | Tara Sesha and Nan Yu (OpenAI)
OpenAI product leaders Tara Seshan and Nan Yu join Claire Vo at the Lenny and Friends Summit to discuss the realities of building AI products in a fast-moving landscape. They explore why shipping imperfect features is essential, how to structure single vs. multi-agent architectures, bridging the gap with research teams through evals, and why voice and self-driving interfaces will define the next wave of software.
Key takeaways
- Working with AI research teams requires a different toolkit than standard engineering, focusing on user failure modes instead of sprint tickets. Read more →
- Tara Seshan recommends that AI product teams plan only two to three months ahead instead of writing multi-year roadmaps. Read more →
- Nan Yu argues that multi-agent systems fail when they exceed human cognitive limits, leading users to build a single "chief of staff" meta-agent to manage the sprawl. Read more →
- Software onboarding faces a universal bottleneck: the empty input box, where users open ChatGPT or vertical tools and do not know what to ask. Read more →
- An agent that completes 99% of a workflow and drops the final step creates more user frustration than a tool that never started. Read more →
5 articles from this episode
Why OpenAI Plans Only 60 to 90 Days Ahead
Tara Seshan explains why AI product roadmaps fail beyond 90 days, comparing OpenAI's rapid cycles to Stripe's stable fintech market.
Designing AI Agents: Single Identity vs. Multi-Agent Systems
OpenAI product leaders debate single vs multi-agent systems, cognitive load limits, and enterprise data security.
OpenAI Product Leaders on Self-Driving Software and Voice
OpenAI leaders Tara Seshan and Nan Yu explain why self-driving software and voice interfaces will replace the empty input box.
Why OpenAI Uses Computer Use as an Agentic Fallback Layer
OpenAI product leaders explain why 99% task completion fails users and how platform layering solves the agent last-mile problem.
More Lenny's Podcast episodes
- Roles aren't converging—they're expanding | Tamar Yehoshua (Atlassian CPO)Oct 4 · 5 articles
- Context is now the product: Product leadership when software can build itself | Karri SaarinenOct 4 · 5 articles
- The rise of HI-ICs | Elena Verna (Lovable)Oct 4 · 5 articles
- Why Claude can’t be your PM (yet) | Anthropic CPO PanelOct 4 · 5 articles