Cal Newport looks back at a very specific piece of 2025 tech optimism: the claim that “AI agents” would start “joining the workforce” and materially changing business output. The story he tells is familiar from other hype cycles: a plausible-sounding extrapolation from real progress (coding agents getting better at multi-step work) that didn’t generalize the way executives expected.
He points to how rough the first wave of “do things on the web for you” products were in practice. In one example from his reporting, an agent spends a long time failing at something mundane: choosing a value from a drop-down menu on a real estate site. Newport also highlights skepticism from critics like Gary Marcus (“clumsy tools on top of clumsy tools”) and notes Andrej Karpathy’s framing that this is better understood as a “Decade of the Agent,” not an overnight transition.
The takeaway isn’t “ignore AI,” but “stop reacting to vibes.” Newport argues 2026 should be the year we demand concrete evidence and look closely at the real (already-present) impacts of the tools that actually work, instead of treating ambitious predictions as inevitable.