The Clever Nest
Posts
Kevin Weil's Interview - 1

Kevin Weil's Interview - 1

Aaron Samuel
May 14, 2025

Hey clever folks,

Kevin Weil, Chief Product Officer at OpenAI, just shared 12 powerful lessons on the future of AI, from how OpenAI ships products to what startups should (and shouldn’t) build.

Here’s Part 1 of our full breakdown:

1. “The AI You’re Using Today Is the Worst You’ll Ever Use Again.”

It sounds rude, but it is true. AI is evolving faster than any tech in history.

That changes how product teams operate:

You can’t wait for the perfect model.
You build now and improve as models get better.

Weil calls this model maximalism: embrace imperfection today because tomorrow’s model will outperform it. (Some say it’s confidence. Some say it’s reality.)

2. Evals = The Secret Weapon for AI Product Teams

Think of evals like unit tests for LLMs.

To build trustable AI features, you need custom evals that test performance for your use case (e.g., coding, research, UI writing).

Why it matters:

The more accurate the eval → the more ambitious the product can be.

3. How OpenAI Ships: Plan Light, Iterate in Public

Forget heavy roadmaps. OpenAI:

Holds lightweight quarterly plans
Builds bottom-up (teams ship without exec approval)
Prioritizes early launches + rapid iteration

This mindset reduces internal friction and increases learning speed. Mistakes are expected and baked into the process. (Learning on the ground gets real here)

4. Why Chat Is Still the Best AI Interface

Weil is all-in on chat as the core UX for LLMs.

Why?

It’s the most flexible, universal communication method humans have.

– Can handle structured tasks (like filling out a form)
– Can handle ambiguous prompts (like brainstorming or summarizing a doc)

Sure, UIs are great for focused tools. But chat is the fallback for everything else.

5. Where OpenAI Won’t Build And Where Startups Should

OpenAI isn’t trying to dominate every vertical.

They’re focused on core models and infrastructure. Leaving huge whitespace for startups to:

Solve niche problems in healthcare, law, logistics, etc.
Use private data to fine-tune models.
Build tools, apps, and systems on top of OpenAI’s APIs

Their goal? Power the ecosystem, not compete with it. (They are like the ocean where all the other marine life come and thrive.)

6. OpenAI’s Internal Use of AI (And Where They Fall Short)

Even at OpenAI, people use ChatGPT for:

Summarizing docs
Writing specs
Creating evals

But Weil admits they’re not using it as deeply as they should.

He wants more teams to “vibe coding” rapid prototyping alongside AI. One internal tool was built entirely this way by the Chief People Officer.