A new study argues agentic AI's ground rules are already being set - and users aren't in the room.
Researchers compared agentic AI to four earlier cases of software acting on users' behalf: ad blockers, platform recommenders, robo-advisors, and spam filters. They found that the question of whose interests these systems serve got settled through technical choices like APIs, protocol governance, and default settings, rather than open debate. Spam filtering is their sharpest example: inbox quality improved substantially, but users' collective ability to contest how filtering worked collapsed. The paper calls this pattern depoliticization - dressing political tradeoffs as neutral engineering.
The authors trace agentic AI's governance already consolidating around the Model Context Protocol and the Agentic AI Foundation, an industry-run body making calls about what agents can do before most users have weighed in. That mirrors the early, unsettled phase of the four prior cases - except this time, the paper argues, the decisions haven't hardened yet.
History suggests today's default settings become tomorrow's unchangeable infrastructure, so this window for input may not stay open for long.