The more we learn about Android Halo, the more worried I become about Android's future

The Android faithful aren’t new to a crisis of confidence, in no small part because Google’s continued AI-ification of our favorite mobile OS has left them feeling unsure. The drip-fed news about the upcoming Android Halo feature suggests Google is moving full steam ahead to revamp our smartphones into AI-first platforms at their core.

While I’m somewhat optimistic that Google’s Halo feature is taking the right steps to help us keep tabs on our little AI helpers, what we’ve learned (and what we still don’t know) has only made me more concerned about Android’s future.

Halo is the missing piece in Android’s AI ambitions

If you missed the announcement at Google I/O, Android Halo is an upcoming feature that serves as a small hub for checking on AI agents running in the background on your phone. The specifics are still rather vague, but Halo will let agents provide updates, ask questions, and surface results right in your status bar.

In a nutshell, it’s the visual and functional bridge between the user and an emerging class of agentic AI tools — like Google’s own Gemini Spark — that we’re all supposed to be running in the near future. Halo aims to at least partially solve the problems of staying in sync with and keeping control over numerous agents.

Halo is just part of a suite of AI-focused improvements coming to Android in the near future. An updated Privacy Dashboard promises to give us more insights into what AI has been up to on our phones, while the experimental AppFunctions API will let our agents hook into our favorite apps, giving them many more tools to work with.

So, that covers the core parts of a modern AI system: agents like Spark, the apps and tools they can run, security checks, and a hub to keep tabs on what’s going on. Android looks increasingly well-set up for an agentic future, but there’s still a lot we don’t know about how this will all work in practice.

AI tightens Google’s grip on Android

First of all, if you’re already irritated by prompts to summarize your documents or help write an email, Halo might already be causing you to grind your teeth. Hopefully, it won’t be intrusive if you haven’t set it up, but even for casual AI users, a status-bar space taken up by the little Halo icon is a constant reminder to engage. And you can bet your last dollar that Google would rather you use its AI tools rather than anyone else’s.

If you want to get started with agentic work via Google’s ecosystem, the jumping-off point is Gemini Spark. However, it’s only available to Google AI Ultra subscribers at a whopping $99.99 per month. Maybe it’ll eventually come to more affordable plans, but increasingly aggressive usage caps could put a damper on how useful this tool ends up being.

The good news is that Halo is designed, at least in principle, to work with “other supported agents” that may be cheaper or have more generous usage than Spark. But exactly what that means in practice remains a great unknown. There’s no documented mechanism for allowing users to select third-party agents, or for specifying what development, security, or contractual hoops they’ll have to jump through to make the cut.

And that leads us to the other important aspect: verified tools. Gemini supports third-party Model Context Protocol (MCP) integrations with Canva, OpenTable, and Instacart, with Adobe and Spotify slated for the near future. That’s a pretty small list, and whether Google plans to pre-approve all future partners or opens up support for the MCP tools of our choice remains to be seen.

Of course, Google’s AI ecosystem is already heavily integrated with Gmail, Calendar, and Maps. Spark will be the agent that works out of the box with your pre-installed Android apps, but will the company be as forthcoming about letting third-party agents tap into these same features? If you thought the increasingly fortified garden of the Play Store was powerful leverage over the Android ecosystem, controlling the AI agents and gateway gives Google an even longer lever.

And this is all without addressing perhaps the most important questions around user privacy. Granted, Google aims to make it easier to see what agents have been running and wants permission-based toggles to prevent abuse. But none of this addresses the fundamental question of how much personal and potentially sensitive data these agents will have access to, and what recourse we have if we want to say “no thank you.”

Will it even be possible for users to wipe an agent’s memory about the calendar or email details they’ve read? How long is this session data retained, in what jurisdictions, and what opt-out mechanisms will there be? Can these same AI tools run offline, or are we destined to send more and more of our private information to some unknown data center?

The coming months will be decisive. If Google follows through on its architectural promises by enabling seamless third-party agent selection, publishing clear developer guidelines, and implementing granular privacy controls, Android Halo could become the definitive platform for mobile agentic AI.

If those promises remain on paper rather than in practice, Halo risks becoming another showcase for Google’s proprietary stack wrapped in the language of openness.