We killed your startup, lol—what to build when Sam Altman keeps shipping better models

Source: Nate’s Newsletter Date: 2025-04-22 URL: https://natesnewsletter.substack.com/p/we-killed-your-startup-lolwhat-to

Summary

The article confronts the model-obsolescence problem directly: OpenAI and others routinely ship capability improvements that kill products built on yesterday’s limitations. The author references an Altman slide from 2023 that acknowledged the pattern explicitly, then asks the harder question Altman left open — how do you build something defensible when the rules keep changing in one direction? The piece frames the search for durable advantage as the central strategic challenge for AI-era builders.

Implications

• Defensibility is the defining question for AI-native product strategy. Startups competing on raw model capability are in a structurally weak position; the search for moats points toward distribution, proprietary data, workflow depth, and switching costs rather than model quality.
• Feeds the build-vs-integrate thread. The pattern the author identifies — foundation labs consuming the commodity layer — is accelerating, which means the viable product surface is narrowing toward things labs won’t build (vertical depth, trust-sensitive workflows, physical-world integrations).
• Relevant to hiring and resourcing decisions. Teams should be auditing their product roadmaps for features that are one GPT release away from being native functionality, and shifting investment toward layers that compound rather than depreciate.