A new MIT paper tracked software developers before and after they adopted AI tools, finding a dramatic rise in low‑level activity. Coders created or edited almost 300% more files, yet the surge fell to 150% for reviewable work and shrank to roughly 30% for full software releases. The gap highlights how speed gains often stall at human bottlenecks.

The study also examined downstream impact, noting that a surge in mobile‑app releases did not translate into higher download numbers. At the same time, Uber disclosed it burned through its entire 2026 AI budget in a single quarter, prompting a shift to cheaper models. Parallel research showed that mixing open‑source AI agents with premium advisors cuts costs while preserving output quality.

Industry observers argue that current organisational structures blunt AI’s productivity promise. Historical analogues, such as electricity’s delayed impact until factories rewired workstations, suggest real gains will emerge when firms build AI‑centric processes rather than graft tools onto legacy workflows. Companies built around AI, like Anthropic and OpenAI, already enjoy outsized revenue and productivity growth.