AI coding is 10–20x faster for the first 1–2 months, then complexity drags it back down
LadyAshBorg · x · 2026-11-03
SirBlueBorg reports from firsthand project experience that AI coding delivers 10–20x speedups for the first month or two, then complexity kicks in—forcing a choice between shipping brittle code or slowing down to actually understand what's built. He cites RamaswmySridhar's data points: Epoch estimates fixed AI performance gets 13x cheaper per year over 4 years; Anthropic's 100k-line C compiler experiment cost $20k in API spend ($200 per kLOC) but was buggy and not comparable to a real compiler; an early-2026 Microsoft study estimates 25% more merged PRs.
The core argument leans on the steam-era and Intel foundry analogies: dropping new power into old working methods changes little until the work itself is reorganized. Today every change still passes through one human's head—AI made writing code cheap but left that bottleneck untouched, and the microservices-era lesson of small, well-contained contracts may be the way to make agent coding actually scale.
More from coding & agent
- FOCUS: PhD project tackles credit assignment for long-horizon agents with active critic learning — EliasEskin · 2026-11-03
- AI feature usage is up but profit is down: audit these five cost sinks first — djgoel123 · 2026-11-03
- Message boards vs repo files for agent memory: what happens to decisions after week two — jordimr · 2026-11-03
- Local GLM 5.3 Flash finishes the same coding task 4.2x faster than cloud (10:24 vs 43:48) — hyudryu · 2026-11-03
- Dev floats idea of using agents to optimize Unreal Engine's source code — gajesh · 2026-11-03
- CleanWeb v2.8.4: SPA state parsing 95% faster, official LangChain/LlamaIndex packages — EstablishmentTough18 · 2026-11-03