GPT‑5.5 shows that improvements happen across three linked layers — model, app, and tool harness — and that combining modest gains at each layer produces outsized practical capability. When top labs deliver better desktop apps (Codex), website‑gated Pro tiers, and more powerful image/code toolchains, switching costs and vendor control rise even if any single model advance seems incremental.
— This convergence concentrates practical AI power in a few vendors and shapes who benefits from automation, so policymakers and competitors should focus on apps and harnesses as much as model capabilities.
Ethan Mollick
2026.04.23
100% relevant
OpenAI’s GPT‑5.5 Pro (website‑only access), the Codex desktop app, GPT‑imagegen‑2, and the reported speed improvement from 33 to 20 minutes for a complex coding task.
← Back to All Ideas