Inside OpenAI

Logan Kilpatrick, head of developer relations at OpenAI and former Apple machine learning engineer, talks with host Lenny Rachitsky about how OpenAI operates, the rise of GPTs and the APIs, and practical advice for building products on top of AI. The conversation covers prompt engineering, where startups can avoid being squeezed out by OpenAI, and the hiring traits OpenAI values.

01Key learnings

• Products built on general-purpose AI should either be radically better than ChatGPT in specific ways or focus on a narrow vertical where domain knowledge matters, since OpenAI will keep moving into general use cases.
• Give models rich context about who you are, your goals, and your audience; Logan argues context is the main lever for getting specific, non-generic output.
• Treat prompt engineering like briefing a capable human with no background: explain the situation, the audience, and what good looks like instead of asking a bare question.
• Package recurring workflows into custom GPTs with uploaded files and instructions so non-developers can share a domain-specific assistant instead of passing around a half-finished chat link.
• Use embeddings to ground answers in your own corpus of content, such as documentation or transcripts, so responses reference facts you actually have.
• Look beyond chat interfaces: Logan sees the biggest opportunity in experiences that fit the task, such as canvases, direct querying of data, or agents that work in the background and report back.
• Plan for AI becoming normal quickly rather than expecting one model release to change everything; the same problems remain and the tools just get more effective at solving them.
• When hiring, Logan prioritizes high agency and urgency, looking for people who act on customer problems without waiting for consensus across many departments.

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.