Android Capability Protocol (ACP)

This is a purely conceptual thought experiment shared for discussion and inspiration only. No implementation, open-source project, or further development is planned. The ideas are offered to spark conversation among enterprise leaders, Android platform teams, and AI innovators.

Introduction

In 2026, on-device AI has become remarkably capable, yet most enterprise professionals still lose dozens of minutes every day to manual copy-paste rituals between their mobile apps. Notes live in one silo, tasks in another, customer records in yet another. The promise of truly agentic AI — an assistant that can reason, act, and orchestrate across your business tools — remains frustratingly out of reach.

The conceptual Android Capability Protocol (ACP) offers a different path forward. It imagines a simple, secure, Android-native standard that lets any authorized AI assistant discover and invoke structured capabilities inside any enterprise app, all while keeping the app owner fully in control of data, permissions, and confirmation flows.

By bridging the gap between powerful language models and the apps that hold our most valuable business information, ACP could unlock significant productivity gains, reduce compliance risk, and lay the foundation for an entirely new model of enterprise mobile software: one where the AI assistant becomes the primary interface and traditional apps evolve into optional, composable capability modules.

What follows is not a product roadmap but a complete conceptual vision beginning with a real-world day in the life of an account manager and moving into the strategic and technical reasoning behind this approach.

A Day in the Life: Hasini’s World Today and What Tomorrow Could Feel Like

Hasini is a Senior Account Manager at a growing SaaS company. She manages a portfolio of enterprise clients worth several million dollars in recurring revenue and spends most of her day on the road or in hybrid meetings. Like thousands of professionals just like her, Hasini’s primary tools on her Android phone are a secure Account Notes application for capturing client meeting details and her company’s Salesforce-based CRM mobile app for tracking opportunities and action items.

On a typical Tuesday morning in March 2026, Hasini’s day begins with a familiar frustration. While waiting for her first coffee, she receives a message from the Vice President at Acme Corp: “Can you send the updated project timeline and confirm the next architecture review slot?” Hasini remembers discussing these exact points during last month’s kickoff meeting. She opens her Account Notes app, searches for “Acme Corp enterprise deal kickoff,” scrolls through pages of detailed notes, screenshots the relevant sections, switches to the CRM, opens the Opportunity record, pastes the screenshot into a comment, and then manually creates four separate follow-up tasks — one for proposal revision, one for scheduling the architecture review, one for the security questionnaire, and one for sharing the timeline. Fourteen minutes later she is done, but already behind schedule and mentally exhausted from the context-switching.

Later that morning, while preparing for a client demo, she needs last month’s negotiation notes. Another search, another round of copying and pasting. During the afternoon pipeline review with her manager, she spends half the call flipping between apps instead of discussing strategy. By the end of the day, Hasini has lost nearly an hour to manual data movement, knows she may have missed logging two smaller action items, and feels the quiet anxiety that comes from knowing her compliance records are incomplete.

Now imagine the same Tuesday with the conceptual ACP in place.

Hasini opens her preferred AI assistant app… any compliant enterprise-grade host , and simply says: “Find my meeting notes about the Acme Corp enterprise deal kickoff and create follow-up tasks in the CRM: proposal revision due Friday, schedule architecture review, confirm security questionnaire, and share updated timeline with the customer.”

Twenty seconds later the assistant replies: “Done. I located the relevant notes in Account Notes, extracted the key points, and created the four tasks in your Salesforce CRM. The owning app asked for your confirmation before creating them — you approved in a tap. Here is a clean summary of the original note for reference. Anything you’d like to adjust?”

Hasini’s entire workflow is complete before her coffee goes cold. No app switching. No retyping. Full audit trail automatically recorded in both Account Notes and the CRM. She reclaims nearly an hour of her day for strategic thinking and client conversations. Across a 20-person account team, that single change could free up thousands of productive hours per quarter while dramatically improving data accuracy and compliance posture.

This is not science fiction. It is the practical productivity leap that a thoughtful, secure interoperability protocol could deliver.

The Fragmented Reality of Enterprise Mobile AI Today

Enterprise organizations have invested heavily in mobile productivity tools and on-device AI. Yet the experience remains disappointingly fragmented. Each business app, whether built by a vendor or internally, operates as its own closed kingdom. AI assistants can chat brilliantly, summarize documents, or generate ideas, but they cannot reliably reach into the systems of record where the actual work happens.

The result is a hidden tax on productivity. Sales teams lose time searching for notes. Account managers duplicate effort entering tasks. Compliance officers worry about incomplete audit trails. And CIOs watch their substantial investments in AI and mobile infrastructure deliver far less business value than promised.

The deeper issue is architectural. Most current AI integrations are either app-specific (each vendor builds its own fragile assistant) or rely on brittle techniques for data acquisition/aggregation. Neither approach scales across an enterprise portfolio, respects data governance boundaries, or survives platform updates. What enterprises actually need is a common, secure language that any AI assistant can speak with any business app without compromising control or compliance.

Envisioning the Agentic Enterprise Mobile Experience

Picture an enterprise where the AI assistant becomes the universal conductor rather than just another chat interface. A professional like Hasini could focus entirely on strategy and relationships while the assistant handles the mechanical orchestration across her tools… always with explicit permission, always within the governance boundaries set by each application owner.

This is not about replacing apps. It is about liberating the structured capabilities inside those apps so they can be composed intelligently. Reading notes becomes effortless. Creating tasks becomes automatic. Updating records happens only after the owning application confirms the action. The assistant reasons, plans, and executes but the enterprise apps remain the authoritative source of truth.

The strategic prize is substantial: higher employee productivity, fewer errors, stronger compliance, and a genuine competitive advantage in how quickly teams can turn insight into action.

A Conceptual Bridge: The Android Capability Protocol

To make this vision real, we need a missing piece… a clean, open, Android-native protocol that standardizes how apps expose their capabilities to AI assistants. That is the idea behind the conceptual Android Capability Protocol (ACP).

ACP is not a new cloud service or a centralized registry. It is a lightweight, on-device interoperability layer built entirely on proven Android patterns. Any enterprise app that chooses to participate simply declares a standard capability service. Any AI assistant on the same device can discover those capabilities, request scoped permission once, and invoke them through structured, machine-readable requests. The owning app remains completely in control of what is exposed, who can access it, and whether any action requires explicit user confirmation inside its own trusted interface.

Because everything stays on the device and uses native Android security mechanisms, ACP aligns naturally with enterprise requirements around data residency, auditability, and least-privilege access.

How ACP Would Actually Work in Practice

In Hasini’s transformed Tuesday, the process unfolds transparently and safely. When she gives the instruction, her AI assistant first checks which installed apps have declared ACP support … in this case Account Notes and the CRM. It retrieves a rich description of each app’s available capabilities, including exactly what inputs they accept and what level of permission is required.

For the first use, each provider app surfaces its own simple consent dialog: “Allow this AI assistant to read Account Notes?” and “Allow this AI assistant to create tasks in the CRM?” Hasini approves once; the permissions are granular, revocable at any time, and tied to the specific assistant application.

The assistant then sends precise, structured requests … never vague natural language. It asks Account Notes to search for the Acme Corp kickoff notes. The results come back in clean, schema-described format. The assistant reasons over the content, identifies the four follow-up actions, and issues create-task requests to the CRM.

For the write operation, the CRM app briefly opens its own confirmation screen… the same trusted interface Hasini already knows and trusts, asking her to review and approve the four tasks before they are created. Once confirmed, everything is logged locally in both applications for compliance purposes, and the assistant returns a clear summary.

The entire experience feels like magic to Hasini, yet every safeguard remains intact. The AI orchestrates; the enterprise apps govern.

The Guiding Principles Behind the Conceptual Design

ACP was conceived around a small set of deeply held convictions.

First, provider sovereignty must never be compromised … the application that owns the data must always decide what is exposed and under what conditions.

Second, user consent must be explicit, granular, and revocable.

Third, safety must be graduated: reading notes should be frictionless after initial approval, while any destructive or externally visible action must trigger confirmation inside the owning app.

Fourth, everything must be structured and machine-readable so modern language models can understand and reason reliably.

And finally, the entire system must be auditable by design, producing clear local records that satisfy even the strictest enterprise compliance regimes.

These principles are not theoretical. They are the practical requirements that would emerge when enterprise security and platform teams actually accept in production.

Deeper Technical Foundations for Solution Architects

For teams ready to explore implementation details, the conceptual ACP surface is deliberately minimal and Android-native. Discovery happens through the standard PackageManager. Each participating app exposes a bound service that returns a comprehensive manifest describing its tools, resources, required scopes, and safety characteristics … all expressed in familiar JSON Schema format.

Tool invocations are simple request-response exchanges carrying structured JSON payloads. For high-risk actions, the provider can return a special status indicating that interactive confirmation is required; the assistant then launches a safe PendingIntent that hands control entirely back to the owning app’s UI. Audit records are written locally by the provider, ensuring a tamper-evident trail that lives where the data lives.

Because the design reuses existing Android security primitives … Binder identity verification, scoped permissions, and ContentProviders for read access, it requires no new platform privileges and can be back-ported to earlier Android versions through a lightweight library.

Enterprise-Grade Safeguards: Security, Compliance, and Control

Enterprise adoption demands more than technical elegance; it requires iron-clad governance. Under the conceptual ACP model, sensitive fields are redacted by default. Providers can expose different capability sets based on user role or device-management profile. Every invocation carries the identity of the requesting assistant, allowing organizations to maintain approved-host allow-lists. And because no data ever leaves the device unless the provider explicitly permits an external-effect action, data-residency and sovereignty concerns are addressed at the architectural level.

For regulated industries, the local audit records provide exactly the evidence auditors ask for: who accessed what, when, and whether the action was confirmed by the user inside the trusted application.

The Longer-Term Horizon: From Apps to Capability Modules

Once a standard like ACP exists, something profound begins to shift. Developers start to realize that many enterprise features… summarization, task creation, compliance checks, renewal forecasting do not actually require a full visual application. They can be packaged as lightweight, focused capability modules that exist solely to serve AI assistants.

The assistant becomes the primary interface. Traditional apps continue to provide rich visual experiences where needed (complex editing, dashboards, creative work), but the structured, repetitive, and orchestratable parts of business work quietly migrate to composable capabilities. The result is a more modular, more agile, and ultimately more powerful enterprise software ecosystem.

Comparison with Google’s AppFunctions

Google’s AppFunctions, introduced with Android 16, is an impressive and production-ready implementation of many of these same ideas, optimized especially for Gemini and system-level agents. It validates that the industry clearly recognizes the opportunity.

The conceptual ACP approach differs in two important ways that matter to many enterprise organizations. First, it is deliberately vendor-neutral… designed so, any AI assistant, whether Google’s, a third-party enterprise platform, or even a locally hosted open model, can participate equally. Second, it places even stronger emphasis on provider sovereignty and explicit interactive confirmation flows, ensuring that critical business actions always happen inside the trusted UI of the owning application with full local auditability.

AppFunctions is already shipping and excellent. ACP is offered simply as an alternative thought experiment for organizations that value maximum flexibility, multi-assistant support, and long-term evolution toward capability-centric architecture.

Conclusion: An Invitation to Reimagine Enterprise Mobile AI

The agentic era is no longer coming … it has arrived on our phones. The only question is whether we will let AI remain a clever conversationalist or whether we will give it the secure, structured keys to orchestrate the tools that actually run our businesses.

Conceptual ACP is one possible answer to that question. It respects Android’s security heritage, honors enterprise governance requirements, and opens the door to a future in which the assistant truly becomes the universal interface for work.

This remains a thought experiment.. shared with the intent that it sparks valuable discussion among leaders, architects, and AI builders. What would your organization gain if every professional could reclaim an hour a day through safe, governed AI orchestration? Where else might this pattern apply?

I welcome your perspectives, challenges, and extensions. The future of enterprise mobile productivity may depend on the conversations we start today.