“GPT-5.3-Codex is our first model that was instrumental in creating itself.” Each generation helps build the next, smarter, builds the next faster, smarter still. This is the mechanism behind the exponential timeline: 2022 couldn’t do arithmetic, 2023 passed the bar, 2024 wrote working software, late 2025 the best engineers handed over most coding work, February 2026 new models made everything before feel like a different era.
This recursive loop is why building for obsolescence isn’t just good advice — it’s survival strategy. The elaborate workarounds you build today aren’t deprecated on a normal software timeline; they’re deprecated on an accelerating one. Shumer walked away from his AI for four hours and came back to a completed application — tens of thousands of lines. The AI had opened the app, clicked through features, tested them like a user would, and iterated on its own until satisfied. “I am no longer needed for the actual technical work of my job.”
The job displacement implications are unlike previous automation waves. AI isn’t replacing one specific skill — it’s a general substitute for cognitive work. Amodei predicts 50% of entry-level white-collar jobs eliminated within 1-5 years. This gives urgency to the Don't be the discriminator — be the patron, not the judge argument: the window to define your role as co-creator rather than selector is closing fast, because the implementation gap isn’t a one-time collapse — it keeps collapsing.
Jaya Gupta draws out the strategic consequence: because “AI has also become an input into its own production” — synthetic data, code, evals, compressed experiment cycles — “the rate of change [is] endogenous,” which is “why the shift cycle dropped below the fundraising cycle, and why the half-life of any thesis is shrinking on a curve you can’t extrapolate from history.” A sub-fundraising-cycle shift rate is precisely why Building in AI is running a trading book — you're long some curves, short others, and exposed to correlations that break when they matter rather than underwriting a fixed plan, and it’s the time-acceleration term inside AI strategy is a self-rewriting equation — solving one constraint changes which constraint matters next.