Live with Tim O’Reilly: A Conversation with ChatPRD Founder Claire Vo

The rollout of GPT-5 didn’t go as smoothly as OpenAI hoped: The hype (much of it whipped up by Sam Altman himself) promised much more than what the model could actually deliver. The general consensus is that GPT-5 is an incremental upgrade, not a transformative revision—but that’s not necessarily a bad thing.

If, as I’ve mentioned elsewhere, “OpenAI’s models don’t seem so special anymore,” we may finally be able to set aside the benchmark horse race and start to focus on the real use cases that actually augment workers. Those might employ cutting-edge models like GPT-5 but could also just as easily make use of older, less powerful models (which come with their own advantages, lower token use among them). As founder of product management copilot ChatPRD and host of How I AI, Claire Vo has already been documenting the “specific, practical, and impactful ways” people are using AI to “improve the quality and efficiency of their work,” as she puts it in the introduction to her podcast. So on this episode of Live with Tim O’Reilly, I’m sitting down with her to dive into all things GPT-5: what tasks it’s good at, what its lukewarm reception means for the industry, and where we’re headed next.

Claire was given early access to put GPT-5 through its paces, and she published her results in an excellent episode and follow-up post. Her honest take: GPT-5 is better at some things, worse at others. She’ll fill us in and share some of the feedback she’s gotten from ChatPRD users. But I’m also interested in getting Claire’s perspective as a product leader: How does she balance cost, performance, and other considerations when choosing models? And as the founder of a company that’s somewhat at the mercy of model providers (as Cursor recently discovered), how does she see the relationship between AI companies and the businesses built on top of them evolving in the future?

Finally, what role will agents play in that future?—ChatPRD has a Model Context Protocol (MCP) integration, and Claire is hosting O’Reilly’s October 23 GenAI Superstream, Everyday AI Agents, featuring talks from practitioners already putting them to work to great effect. We’re entering into a new era of computing, where none of the old rules may apply. She recently made a very insightful comment on X that ties some of these threads together: “When agents and tools are the only ones reading your outputs, it makes sense to be as verbose as possible,” Claire argued. But what works best for agents is clearly not the same as what works for ChatPRD’s users. So how do we account for both? It’s definitely worth teasing out. But this is just the framework for our talk. We’ll fill in the details with questions and comments from the audience.

Recommended prep or follow-up:

• Listen to How I AI (podcast)
• Listen to Generative AI in the Real World: “Getting Ahead of the Curve with Claire Vo” (podcast episode)