AI makes coding 10x faster, so why isn't your team shipping 10x faster? Amdahl's law explains
arpit_bhayani · x · 2026-10-23
Arpit Bhayani applies Amdahl's law to explain why AI coding speedups don't translate into team-level shipping gains: overall speedup is capped by the work you didn't accelerate, Speedup = 1 / ((1-p) + p/s). If coding is 30% of the idea-to-production cycle and AI makes it 10x faster, total speedup is only 1.37x; even at 50% of the cycle, 1.8x.
Everything else remains: figuring out what to build, cross-team alignment, code reviews, CI/CD wall time, and debugging AI-generated code nobody wrote line by line. It can get worse than Amdahl predicts — faster code generation loads every downstream stage, so review queues grow and the bottleneck shifts; writing 10x more code with the same review capacity just builds a bigger queue.
His advice: measure the whole path from idea to production, find where work waits, and apply AI there too — review, test generation, flaky test triage, deploy safety checks. "Speeding up the typing was always the easy part."
Related event: AI speeds coding 10x but delivery only 1.4x: Amdahl's Law explains why(2 posts)→
More from coding & agent
- THU-KEG Releases RIDAR to Detect and Intervene in Research Agent Integrity Violations — THU-KEG · 2026-10-23
- One trailing dot let Copilot rewrite its own config: Microsoft patches VS/SSMS RCE (CVE-2026-103572) — wunderwuzzi23 · 2026-10-23
- After LiteLLM mess: delay new packages 24h, but install OS updates instantly — altryne · 2026-10-23
- Dev hooks OpenAI Realtime API to a DIY robot arm: sim-trained skills plus vision for tic-tac-toe — Tarun122 · 2026-10-23
- Same LoRA on every Qwen3.5 size: 9B beats 27B, and nearly all the gain is learning to say no — spilldahill · 2026-10-23
- 200 concurrent LLM agents on a 1GB, $6/month VM — the bottleneck was our own save locks — swarmagent · 2026-10-23