techpolicy.press web signal

Colville: personal AI agents risk a new Big Tech oligopoly

TL;DR

  • The op-ed names Google's Gemini (29 integrated products), Microsoft's Copilot (18), Meta's Muse (10), and Apple Intelligence (29) as leading vectors for agent concentration.
  • Colville warns personal AI agents could enable 'surveillance pricing' and turn the economy into a 'pay-to-play system' tilted toward advertisers and partners.
  • His remedy: open standards, interoperability and data portability, modeled on the UK Open Banking standard, plus a ban on commercial use of sensitive personal data.

Google's Gemini 'Personal Intelligence' is wired into 29 of its products. Microsoft's Copilot sits inside 18. Meta's Muse rides in 10 products including WhatsApp, which the piece pegs at 3 billion monthly users. Apple's Siri and Apple Intelligence span 29 products across the company's hardware and software. xAI's Grok Bot lives in one place: X.

Those numbers come from a TechPolicy.Press op-ed by George Colville arguing that personal AI agents, the Muses, Dots, Duets, Sparks, Copilots and Grok Bots now shipping, are the next concentration battleground. The goal of the category, he writes, is 'to create and monetize a personal assistant that can automate as many daily tasks as possible.' His worry is that one or a handful of providers end up owning it.

Colville's framing is blunt. If agents become the layer through which people transact, they 'would become the arbiters of what people buy and who they buy it from, would select which news they receive, and would shape how they communicate.' Combine that reach with the personal data agents collect and, he argues, 'extreme forms of surveillance pricing would also become possible as monopolists gain access to the unprecedented quantities of personal data.' An agent tuned for advertiser revenue, he writes, 'might prioritize answers and actions that favor partners and advertisers, turning the entire economy into a pay-to-play system.'

His remedy sits in familiar competition-policy territory: mandate open standards, interoperability and data portability, with the UK's Open Banking standard as the reference model, and ban use of sensitive personal data for commercial gain. 'Users, backed by regulation, must be able to ensure that agents are working in their interests,' he writes.

The piece is argument, not evidence. It does not quantify how steep the data feedback loop actually is in agents today, and it points at no pending US, EU or UK enforcement action that would carry the standards it calls for. Two policy voices on our Who's Who list posted the link the same day.

Shared on Bluesky by 2 AI experts