LLenny's Podcast
← All frameworks
StrategyKevin Weil (CPO at OpenAI, ex-Instagram, Twitter)

Model Maximalism

Build at the edge of what models can barely do — the next model will make it sing.

Difficulty
Moderate
Time to result
~months to results
Steps
3
Confidence
90%

AI capability improves roughly every two months, so the model you have today is the worst you'll ever use. Rather than pouring effort into scaffolding that patches current model limitations, build products that sit right at the edge of what models can barely do and let the next model catch up. If your product only just barely works today, that's a signal you're building the right thing.

Origin

Kevin Weil (CPO, OpenAI) describes this as OpenAI's general product mindset and the advice they give to their 3M+ developers.

Core principles

  • 01The AI model you use today is the worst you will ever use for the rest of your life
  • 02Don't over-invest in scaffolding to patch limitations the next model will erase
  • 03A product that 'just barely works' at the edge of model capability is a positive signal, not a warning
  • 04Skate to where the puck is going — build for the capabilities that are almost there

How to run it

  1. 1

    Aim at the current capability edge

    Choose a product idea that sits right at the boundary of what today's models can barely accomplish, rather than something comfortably within reach.

    Watch out If the idea is trivially easy for today's models, the opportunity may already be commoditized.

  2. 2

    Ship the thin version and minimize scaffolding

    Build the product with as little custom error-handling and workaround scaffolding as possible, accepting that it 'just barely works' today.

    Watch out Keep scaffolding only for classes of errors you truly cannot tolerate; strip the rest.

  3. 3

    Keep going and wait out the next model

    Give it another couple of months — when the next, better model lands, the product that barely worked will suddenly perform well.

    Pro tip Treat 'this barely works' as confirmation you're doing something right, not a reason to quit.

In the wild

Bolt / StackBlitz breakthrough

Weil recounts that StackBlitz (Bolt) had been building a failing product for seven years; when Claude Sonnet 3.5 shipped, everything suddenly worked. He notes the same pattern shows up constantly across YC companies as models update every few months.

A long-stuck product became viable overnight because the underlying model crossed the needed capability threshold.

Common mistakes

Over-building scaffolding around today's limits

Investing heavily in workarounds for weaknesses that the next model release will simply eliminate wastes effort.

Abandoning an edge product too early

Quitting because the product only barely works misreads the strongest signal that you're on the right track.

Is it for you?

Best for

AI founders and product teams deciding how ambitious to be given rapid model improvement

Not ideal for

Regulated or safety-critical use cases where errors are unacceptable and you cannot ship something that only barely works

From the transcript

The AI models that you're using today is the worst AI model you will ever use for the rest of your life

00:00

the product that you're building is kind of right on the edge of the capabilities of the models, keep going

00:30

in two months there's going to be a better model and it's going to blow away whatever you know the current set of limitations are

31:30

From the episode

OpenAI’s CPO on how AI changes must-have skills, moats, coding, startup playbooks, more

Kevin Weil (CPO at OpenAI, ex-Instagram, Twitter)