Where AI Is Headed: 10 Grounded Predictions for 2027

Prediction lists age badly, so these ten stick to trajectories already visible in the data — extrapolation, not prophecy.

By 2027 we expect: agents handling most tier-one support and routine back-office work; AI-native employees becoming a hiring category with 'orchestrates agents' in job descriptions; on-device models handling the majority of everyday queries for latency and privacy; video generation reaching broadcast-usable quality with provenance labeling standard; open models permanently 6-12 months behind frontier closed ones — close enough for most business use; regulation converging on disclosure and liability rather than capability bans; and the first AI-discovered drug candidates reaching late-stage trials.

The meta-prediction: the gap that matters will not be between AI models, but between organizations that redesigned workflows around AI and those that sprinkled it on top. The tools are ready; operating models are the bottleneck.