The Specific Event

On Thursday, OpenAI rolled out support for ChatGPT to control Apple's iMessage service directly. Bloomberg reported the development without fanfare, framing it primarily as a privacy concern for Apple. That framing is understandable but incomplete. The more consequential issue is not whether OpenAI can read your messages. It is what happens when an AI system becomes the coordination layer sitting between a user and their communication infrastructure.

This is a structural change in how platform coordination works, and organizational theory has not fully caught up to what that means.

When the Agent Becomes the Interface

Classical coordination theory assumes that the parties doing the coordinating have direct access to the environment they are operating in. Markets coordinate through price signals that participants read and respond to. Hierarchies coordinate through authority structures that individuals navigate. Platforms, as I have argued in my dissertation work, represent a distinct mechanism where competence develops endogenously through participation in algorithmically-mediated environments (Kellogg, Valentine, and Christin, 2020).

ChatGPT controlling iMessage introduces a layer that none of these frameworks fully account for. The user is no longer participating in the platform environment directly. An AI agent is doing the participating on their behalf. This is not a small variation on existing platform coordination. It is a category shift. The user's folk theories about how iMessage works, their adaptive expertise about when to respond, how to frame a message, or what communication norms apply, become irrelevant if an agent is executing those decisions autonomously (Hancock, Naaman, and Levy, 2020).

The Awareness-Capability Gap Gets Inverted

One of the central puzzles in my ALC framework research is the awareness-capability gap: workers can know that algorithms govern their outcomes without knowing how to respond effectively to those algorithms. Algorithmic awareness does not produce algorithmic competence (Gagrain, Naab, and Grub, 2024). ChatGPT's iMessage integration creates what I would call the inverse condition.

Here, the user does not need awareness of the underlying platform mechanics because the agent abstracts them away entirely. But this produces a different and potentially more serious problem. When the agent makes a coordination error, the user has no schema for diagnosing what went wrong. They have been systematically excluded from the feedback loop that would have built that schema in the first place. Sundar (2020) describes this dynamic as machine agency displacement, where delegation to automated systems erodes the human's capacity for meaningful oversight over time.

This is not hypothetical. If ChatGPT sends an iMessage on your behalf and the recipient responds with confusion or offense, you lack the structural understanding of what the agent did or why. You have traded procedural participation for convenience, and in doing so, you have made yourself less capable of recovering from failure.

Organizational Consequences Are Already Visible

The organizational theory implications here extend well beyond individual users. Tanium's concurrent news this week - a $9 billion cybersecurity firm bringing its cofounder back as CEO amid AI-related upheaval - signals that enterprise organizations are recognizing a governance gap in how AI agents interact with communication and security infrastructure. The leadership change at Tanium is at least partly a response to the same structural condition: organizations that delegated coordination to AI systems are now scrambling to reestablish oversight capacity they allowed to atrophy.

This is precisely the condition Rahman (2021) describes in the context of platform dependence: once an organization or individual structures their coordination around an external algorithmic system, exit becomes costly not because leaving is technically difficult, but because the competencies required to function without the system have eroded. ChatGPT controlling iMessage at scale would accelerate that process considerably.

What This Means for Platform Theory

The ALC framework I am developing treats platform coordination as a site where competence is built through direct participation. The agent-mediated model that OpenAI is now deploying challenges one of the core premises of that framework. If users increasingly coordinate through AI intermediaries rather than through platforms directly, the endogenous competence development mechanism breaks down. Platform workers and communicators would no longer be developing structural schemas through experience. They would be outsourcing the experience itself.

Whether that produces better coordination outcomes in the short term is an empirical question. What is theoretically clear is that it produces more fragile coordination over time. Routine expertise, which Hatano and Inagaki (1986) distinguish from adaptive expertise, is exactly what agent-mediated systems provide. They optimize for predictable conditions and degrade sharply in novel ones. The iMessage integration is a convenient feature. It is also, at scale, a coordination liability.