Convergence
Six operators independently argue that the dominant strategic error in AI product work is over-investing in scaffolding for the model that ships today. The next model lands in 3-6 months and eats most of the workarounds — the team that built lightly, kept the goal abstract, and re-baselined often is in front. The current model is "the worst it will ever be."
Operators
- Sherwin Wu (Head of Eng, OpenAI API) — Build for the model six months out — the current model will eat your scaffolding. Frame: customer asks for V1 scaffolding are valid but depreciating; do the asks, but don't bet the surface on them.
- Boris Cherny (Claude Code, Anthropic) — Build for the model six months out, not the one that ships today and Give the model tools and a goal; do not hard-code the workflow. Frame: give the model a goal and tools, do not hard-code the workflow; six-month forecast horizon for surface decisions.
- Cat Wu (Head of Product, Claude Code) — Build products at the edge of what does not yet work and When a new model lands, re-read the system prompt and remove crutches. Frame: ship at the edge of what doesn't quite work yet; on every model upgrade, re-read the system prompt and remove crutches.
- Ethan Mollick — Pin AI workflows to capabilities you can re-baseline quarterly, not to one model snapshot. Frame: pin to capabilities you can re-baseline quarterly; don't pin to a snapshot.
- Benjamin Mann (Anthropic) — Use new tools as new tools, not as old tools — be ambitious and retry from scratch. Frame: don't use new tools as old tools — retry the original ambition with the new ceiling.
- Simon Willison — November 2025 was the qualitative threshold — coding agents now almost always do what you tell them. Frame (counter-anchor): identifies a moment (Nov 2025) when capability crossed a threshold; reinforces "the floor moves" thesis.
Variation
- Sherwin and Cat operate from inside model labs — their stake is "stop asking the lab for V1 scaffolding."
- Boris extends to a design rule: keep the workflow abstract, keep tools and goals concrete.
- Mollick frames it as a measurement discipline: re-baseline, don't pin.
- Mann frames it as ambition discipline: don't replicate the old workflow at higher speed.
- The shared move is restraint on scaffolding; the variation is whether the operator emphasises the measurement loop (Mollick), the design rule (Boris), the ambition reset (Mann), or the surface debt (Sherwin/Cat).
Implication
Treat current-model scaffolding as a depreciating asset. Ship the customer fix, but don't anchor your roadmap or org structure to it. Add a quarterly re-baseline to your operating cadence. Hire builders who can rewrite from scratch when the ceiling moves.
Sources
- ins_build-for-next-model-not-current — Sherwin Wu
- ins_build-for-model-six-months-out — Boris Cherny
- ins_build-products-that-dont-yet-work — Cat Wu
- ins_remove-features-as-models-improve — Cat Wu
- ins_rebaseline-quarterly-not-pin-to-snapshot — Ethan Mollick
- ins_dont-box-the-model-in — Boris Cherny
- ins_use-new-tools-as-new-tools — Benjamin Mann
- ins_november-2025-coding-inflection — Simon Willison (anchor evidence)