The Information Machine

New MIT study. Code volume surges by 300%, but output increases by only 30%: The AI dividend meets an awkward reality

Rohan Paul Twitter · Rohan Paul (@rohanpaul_ai) · 2026-06-07

A new MIT study finds that AI coding agents increased code volume by 300% and commits by 180% but raised actual software releases by only 30%, because software delivery is bottlenecked by weak links that faster code generation cannot eliminate.

Open original ↗

Appears in

Extraction

Topics: ai-codingsoftware-productivityllm-agentsresearch

Claims

  • AI coding tools have increased code volume by 300% but improved actual software output by only 30%.
  • Autonomous AI coding agents raised commit counts by 180% while software releases rose by only 30%.
  • Software production is governed by weak links, so accelerating coding alone does not proportionally increase end-to-end delivery.
  • The productivity dividend from AI coding assistance is substantially smaller than raw activity metrics suggest.

Key quotes

Code volume surges by 300%, but output increases by only 30%: The AI dividend meets an awkward reality
Autonomous AI coding agents raised commits by 180%, but releases rose only 30%.
software production has weak links, so faster [coding does not translate proportionally to faster delivery]