Topic
AI agents building and shipping software end to end without human coding in the loop.
LinearB's 2026 benchmark finds only 32.7% of AI-assisted pull requests accepted within 30 days, against 84.4% of manual ones. The figures are associations, but they put review cost in the queue and in repeat rounds, where faster diff reading helps little.
Reality
One operator ran spec-driven development at full BMAD weight for four weeks on an autonomous coding repo. The planning stage produced 1,782 files and 16 MB of artifacts, and a single story burned around thirty million tokens.
Reality
Orchid's new controls detect when an agent drifts from its declared purpose and then cut its access, and both halves depend on inventory and specification work the enterprise has to finish first.
Reality
Three sequential runs with no task produced 77 turns, $6.96 in billing and one real commit. The scratch directory was swept each time, and the state that survived is what set the agenda.
Reality
A VC that reviewed hundreds of AI productivity tools in a year says most are headed for the graveyard, and capital should rotate to drug discovery, defense and physical AI.
Reality
A post-race audit of PricePulse, built unsupervised by Claude, found features that mostly worked and an organisation that did not. The failures were startup failures, not model failures.
Reality