Build for the model six months from now, not the model of today

Boris Cherny repeats this as Anthropic’s core product philosophy: “We don’t build for the model of today. We build for the model six months from now.” He frames it as advice to LLM founders: identify the frontier where the model is not very good today, because it’s going to get good at it.

The proof is in Claude Code’s own history. When Boris started building it in September 2024 with Sonnet 3.5, the model couldn’t really code — “even in February when we GA’d, it wrote maybe like 10% of my code.” The team bet on future capability and was rewarded: “our bets paid off and it got good at the thing that we thought it was going to get good at, because it wasn’t obvious.” This connects to Scaffolding is tech debt against the next model — the bitter lesson applied to product building — the bitter lesson means you should let the model solve problems rather than engineering around its limitations.

The competitive implication is stark: “Otherwise what will happen is you spend a bunch of work, you find PMF for the product right now, and then you’re just going to get leapfrogged by someone else, because they’re building for the next model.” This extends Model-market fit comes before product-market fit — without it, no amount of product excellence drives adoption — model-market fit isn’t static; it’s a moving target, and the founders who anticipate where it’s heading build more durable products. The insight also reinforces Frontier companies absorb every useful agentic pattern into their products: if you build scaffolding for today’s limitations, the frontier company will absorb that pattern into the model itself, making your product layer redundant.

Thariq Shihipar’s Anthropic-internal framing explains why this bet is sound rather than speculative: Models are grown, not designed — treat a new model class as biology, not physics — capability isn’t hand-designed to a spec, it’s grown and then discovered, which means the frontier a model “isn’t very good today” is a real, probeable gap, not a guess. And The model already knows the answer — it just can't reach it without the right tool sharpens what “building for the model” actually means in practice: often the future capability is already latent in the current model, waiting on the right tool or harness to surface it, rather than requiring a future training run at all.