Product Overhang
Product overhang is the gap between what a model can already do, or will soon be able to do, and what existing products have captured in usable workflows.
Key points
- Boris Cherny says Anthropic used the term heavily inside Anthropic Labs while building Claude Code: the model was approaching a capability level where it could write whole chunks of code, while mainstream products still focused on IDE type-ahead [src-054].
- Claude Code was deliberately built before the model was fully ready. Boris says it was barely usable for roughly six months and only reached strong growth after Opus 4, then kept inflecting with later models [src-054].
- The strategic bet is to build the product shape that the next model will make useful, not only the product that today’s model can satisfy reliably [src-054].
- Boris names Claude Design, massively parallel agents, loops/batch, and computer use as current examples of product areas that become much more interesting as models improve [src-054].
- Karpathy gives a Software 3.0 version of product overhang: some applications should no longer exist as conventional app shells because a model can transform the raw input directly into the desired output, as in the MenuGen example with Gemini/Nano Banana [src-055].
- This makes product overhang broader than coding tools. It also applies to information processing workflows that could not previously be automated with Software 1.0 code [src-055].
Related entities
Related concepts
- Coding Democratization
- AI-Native Startups
- AI-Native Organizational Process
- Software 3.0
- Agent-Native Infrastructure