Has AI Changed Work Forever? Not Really... | Cal Newport

The gist

1. AI progress slowed in 2025 as pre-training scaling hit diminishing returns, limiting gains to narrow areas rather than broad workplace transformation.
2. Newport reviewed 250 plus real programmer workflows and found no one lets AI run for hours unsupervised; developers test constantly and abandon the AI about 20% of the time.
3. AI coding tools handle repetitive tasks like API integration well but cannot do the mathematical or conceptual work that drives actual AI breakthroughs.
4. Newport says companies focused on code generation because it was the only niche where they found traction after failing in video, general agents, and other areas.