The Great Uncoupling
Something fundamental shifted in enterprise governance this year. The traditional model — where humans own systems, systems process data, and policies govern both — has quietly collapsed. In its place, we're witnessing the emergence of autonomous systems that nobody fully owns, processing data products that nobody fully controls, all while generating outcomes that nobody can fully verify.
This isn't hyperbole. Consider Atlassian's recent revelation: they built an entire AI agent infrastructure platform in just four weeks using their own Rovo agents. The recursive loop is striking — AI building the infrastructure for more AI, with human developers increasingly relegated to orchestration roles rather than direct creation. When the builders become the built, who exactly owns the outcome?
The Multiplication Problem
The ownership crisis becomes acute when you examine the numbers. Organizations deploying AI agents report managing dozens, sometimes hundreds of autonomous systems. Each agent requires its own governance model, its own accountability framework, its own audit trail. Yet traditional ownership models assume a human at the top of every chain.
This multiplication effect extends beyond AI. Data products — those self-contained, reusable data assets that organizations increasingly rely on — present similar challenges. As one industry observer noted, the quality of a data product depends equally on its data and its code. But when that code is generated by AI, maintained by automated systems, and deployed through infrastructure that itself was built by agents, the ownership chain becomes impossibly tangled.
The enterprise architecture community has begun calling this the "90-day problem." In the new software economics, development cycles have compressed so dramatically that traditional governance reviews can't keep pace. By the time you've assigned ownership, defined accountability, and established controls, the system has already evolved twice.
The Accountability Vacuum
The consequences are already visible. The Chartered Institute of Internal Auditors reports that internal control failures have resulted in over £1 billion in fines — and that's just what's been caught and prosecuted. The real number, accounting for undetected failures in increasingly autonomous systems, is likely far higher.
What makes this particularly challenging is that traditional governance assumes clear lines of accountability. When a system fails, we look for an owner. When a decision goes wrong, we seek a responsible party. But in a world where:
- AI agents build and deploy other AI agents
- Data products self-modify based on usage patterns
- Infrastructure provisions itself based on predicted demand
- Code generates code that generates code
...the very concept of ownership becomes philosophical rather than practical.
The False Solution of Registries
The knee-jerk response has been to build registries — comprehensive catalogs of every AI agent, every data product, every automated system. But registries only document existence; they don't establish ownership. Knowing you have 500 AI agents running across your infrastructure doesn't tell you who's responsible when agent #247 makes a million-dollar error.
Worse, the registry approach assumes stability. It works when systems are created, registered, and then remain relatively static. But modern AI agents evolve continuously. They learn, adapt, and modify their behavior based on interactions. Today's registry entry may bear little resemblance to tomorrow's operational reality.
The Governance Inversion
We're witnessing what might be called a governance inversion. Traditionally, governance flowed from the top down — boards set policy, executives implemented it, systems enforced it. But when systems become autonomous, governance must flow from the bottom up. The agents themselves need embedded governance capabilities.
This isn't about making AI "ethical" or "aligned" — those are separate challenges. This is about basic operational governance: Who can modify this system? Who bears liability for its decisions? Who has the authority to shut it down? When an AI agent spins up cloud infrastructure that costs $100,000 per month, who approves that budget?
Some organizations are experimenting with "governance by design" — building ownership and accountability directly into automated systems. Each AI agent carries its own governance metadata: its owner, its authority limits, its accountability chain. But this approach faces its own paradox: Who governs the governance layer?
The Path Forward
The ownership void won't be filled by traditional governance approaches. We need new models that acknowledge three uncomfortable truths:
- Ownership must be dynamic — as systems evolve, ownership and accountability need to evolve with them
- Collective ownership is inevitable — when multiple AI agents collaborate to produce an outcome, responsibility must be shared
- Automated governance is necessary — human-speed governance cannot govern machine-speed operations
The organizations that thrive in this new reality will be those that stop trying to force autonomous systems into human-centric governance models. Instead, they'll build governance that operates at machine speed, adapts to machine evolution, and acknowledges machine autonomy while still maintaining human accountability.
The ownership void is real, growing, and accelerating. But it's not insurmountable. It simply requires us to rethink fundamental assumptions about who — or what — can own, control, and be held accountable for enterprise systems. The future of governance isn't about controlling autonomous systems. It's about creating governance frameworks that are themselves autonomous, adaptive, and accountable.
The question isn't whether machines can own things. It's whether we're ready to govern in a world where ownership itself has become fluid, shared, and increasingly abstract. The organizations that answer this question first will define the next era of enterprise governance.
Sources
- UPDATED! The Ultimate Agentic AI Governance Resource Guide — Enterprise AI Governance (Oliver Patel)
- Rovo Dev Driven Development – How we built a platform in 4 weeks — Atlassian Work Life Blog
- The New Software Economics: Earn the Right to Invest Again, in 90-day Cycles — Architecture & Governance Magazine (Iasa)
- Strategies for Security and Governance in AI with insightsoftware, Informatica, and OpenText — DBTA (Database Trends & Applications)
- The quality of a data product is its data and its code — Andrew Jones (Data Contracts)
- Internal controls failures see firms fined £1bn+ — Board Agenda