Build for the Model Six Months Out
Design your AI product for the model that ships in six months, not the one you have today.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 4
- Confidence
- 90%
Because model capability improves on an exponential, building for today's model leaves you rebuilding constantly. Instead, build for where the model will be in ~6 months. The cost is uncomfortable — your product-market fit looks weak for the first stretch — but when the better model lands, the product suddenly clicks and you hit the ground running.
Origin
Boris Cherny's account of how Claude Code was built at Anthropic; he frames it as advice he gives to startup founders building on AI.
Core principles
- 01Model capability follows an exponential, so extrapolate the line rather than reacting to the present
- 02Early PMF will feel bad because the model isn't ready yet — that discomfort is expected
- 03Concrete capability bets: models get better at tool/computer use and at running unattended for longer
- 04When the next model lands, a product built ahead of it inflects instead of needing a rewrite
How to run it
- 1
Extrapolate the capability line
Look at the trajectory of what the model can do and project ~6 months forward — especially tool use, computer use, and how long it can run unattended.
Pro tip Trace the exponential, not your intuition. Boris predicted crossing 100% AI-written code simply by 'tracing the line.'
Watch out Intuition consistently underestimates exponential curves; expect the projection to feel absurd.
- 2
Design the product the future model unlocks
Build the experience that only works once the model is good enough — e.g. an agent that runs for 30 minutes unattended rather than needing constant hand-holding.
- 3
Tolerate weak early PMF
Accept that for the first ~6 months adoption and fit will look poor because the underlying model hasn't caught up yet.
Watch out Don't kill or heavily pivot the product on early metrics alone if the bet is the model, not the design.
- 4
Ride the inflection
When the stronger model ships, the product suddenly works. Claude Code inflected with Opus 4 and again in November when it crossed 100% of Boris's code.
In the wild
At launch Claude Code was not immediately a hit; it wrote ~20% of Boris's code in February and ~30% in May while he still used other tools. The team had bet on a future model. When Opus 4 and Sonnet 4 arrived, everyone started using it for the first time and growth went exponential.
→ A product that looked mediocre for months became a multi-billion-dollar business once the model it was built for arrived.
Common mistakes
Building for today's model
A product tuned to current capabilities needs constant rework as models improve and never gets the compounding payoff of a well-timed bet.
Abandoning too early on weak PMF
Because the first six months intentionally look bad, founders who judge purely on early traction quit right before the model catches up and the product clicks.
Is it for you?
Best for
Founders and product teams building AI-native products who can absorb a slow first six months
Not ideal for
Teams needing immediate revenue or PMF, or products whose value doesn't depend on model capability
From the transcript
“we bet on building for the model 6 months from now.”
“if you build for the model 6 months out, when that model comes out, you're just going to hit the ground running”
“It's going to be uncomfortable cuz your product market fit won't be very good for the first 6 months.”
From the episode
Head of Claude Code: What happens after coding is solved
Boris Cherny