METR’s latest survey of 350 technical knowledge workers revisits the long‑standing claim that AI boosts productivity. Earlier work found engineers felt a 20 % speedup, yet actual measurements showed a 20 % slowdown. The new study shifts focus from speed to added value, seeking a more realistic picture for businesses and investors today in the tech sector.

METR’s revised questions asked workers to estimate how much more value AI adds to their outputs rather than how fast tasks complete. Respondents reported a 2× increase in value on average, with a third question capping it at 1.6×—an 60% boost. Still, analysts caution these figures are upper bounds given self‑report bias and role variability.

Parallel telemetry from Faros, a software‑development platform, paints a different picture. Data from 22,000 developers shows that while AI drives more code and project starts, downstream stages slow dramatically. Quality drops, and incidents triple, forcing teams to spend extra hours firefighting and eroding overall value for client delivery and stakeholder confidence in the product pipeline.

These findings warn that individual productivity gains may not translate into corporate or economic benefits. As AI tools become mainstream, firms must rethink workflows and quality controls to avoid a productivity–quality trade‑off. The real test will be whether organizations can harness AI’s speed without sacrificing the robustness that underpins market trust for customers and shareholders.