Chapter 3 • Early 2026Observed Reality (as of Nov 2026)Track: spine
Early 2026: Coding Automation (1.5x AI R&D Multiplier & SL3 Security)
Using Agent-1 internally for ML engineering yields a 50% (1.5x) speedup in algorithmic progress; OpenBrain upgrades model weight security from RAND SL2 to SL3.
OpenBrain deploys Agent-1 across its internal research workflows, achieving a 1.5x overall pace of algorithmic progress compared to human researchers alone. Recognizing that stolen weights would hand competitors a 50% R&D boost, leadership upgrades from RAND Security Level 2 (SL2) to SL3.
Observed Reality (as of Nov 2026): Partially Realized. Frontier labs widely use internal coding agents for kernels and evals, with estimated net R&D multipliers of 1.15x–1.45x as of November 2026.
Verification Note: Partially realized as of Nov 2026: internal coding agents are ubiquitous at frontier labs, though end-to-end R&D uplift is closer to 1.25x–1.4x than a full 1.5x.
Frontier labs capture a 1.5x internal R&D speed multiplier 12–18 months before the broader macroeconomy because ML researchers dogfood their own unreleased coding agents.
In the AI 2027 Timelines Forecast, we model the initial 2026 R&D acceleration at 1.5x (growing toward 3x by Jan 2027). Because frontier labs have zero security-procurement lag for their own models and fine-tune specifically on their internal CUDA/Triton/JAX monorepos, internal algorithmic progress speeds up well before Fortune 500 enterprises reorganize.
steelman@ScottAlexander(Scott Alexander (Astral Codex Ten))Astral Codex Ten / AI 2027
Why 3x faster coding only yields 1.5x faster AI research in early 2026: GPU experiment wall-clock time acts as an Amdahl's Law governor.
If a human researcher spends 50% of their week writing/debugging experiment code and 50% waiting for GPU cluster jobs to finish and analyzing loss curves, even infinite coding speed can only double throughput until the AI also designs more sample-efficient experiments.