The Architecture of Agency: Why No Single Actor Can Survive the Transition Alone

In January 2023, the Philips Healthcare plant in Kearney, Nebraska announced it would eliminate 320 positions — assembly workers who built ventilator components now largely produced by automated systems in Suzhou. By February, three things happened at once. Angela Mercado, a 14-year assembler, enrolled in a welding certification. Kearney's labor co-op convened a structured working session where affected workers mapped their collective skills and gaps. And the Nebraska Department of Economic Development activated a rapid-response team with retraining vouchers and employer tax credits.

None of those three actions was sufficient alone. Together, they produced something none could manufacture independently. Eighteen months later, 214 of the original 320 displaced workers had placed into new employment, and the labor co-op had shared its model with co-ops in six other Nebraska towns.

That is the subject here: the architecture of agency that makes transitions survivable — and occasionally generative.

The tension: top-down and bottom-up are both incomplete

The reflexive debate about managing technological disruption splits into two camps. One trusts institutions: write the regulation, fund the program, mandate the standard, and order will follow. The other trusts the grassroots: communities and individuals will self-organize if institutions get out of the way. Both intuitions contain truth. Both, alone, fail.

The institution-first view keeps running into a hard limit. The Preston Model in Lancashire redirected anchor-institution procurement back into the local economy — universities, a college, an NHS trust mapping where their spending actually went. That created a scaffold. What it could not create was the human being willing to step through the door and facilitate a contentious budget meeting. The grassroots-first view hits the opposite wall: a neighborhood pod cannot supply the patient capital a worker cooperative needs to survive its first competitive crisis.

The reframe: durable order grows from the middle

Elinor Ostrom won a Nobel Prize studying commons governance across Maine fisheries, Swiss forests, and Indonesian irrigation systems. The pattern held everywhere she looked: durable systems require multiple autonomy levels — communities adapting rules locally, institutions coordinating without imposing, individuals participating in governance. Robert Putnam's research confirms the complement: trust and resilience emerge through civic networks, not hierarchical command.

Industrial towns in Germany. Smallholder communities in Fiji. Fishing commons in Maine. The contexts are too different to dismiss the recurrence as coincidence — and that regularity is itself the argument. It cuts against the intuition that durable order must be imposed from the top. Again and again, it grows from the middle.

Figure: The Three Protagonists — agency distributed across three levels, each doing work the others cannot, none sufficient alone.

The mechanism: three protagonists, acting in relationship

The architecture has three actors, and they don't operate in sequence. They operate in relationship.

The Individual is the engine of moral agency. Every transformation worth studying began with people who refused to accept existing arrangements as permanent. When a Machine Core can execute a thousand variations of any strategy, the scarce resource is clarity about which variation serves human flourishing. The individual becomes the conscience of systems already too powerful to run without one.

The Community is a learning system. After the Philips announcement, Kearney's co-op acted — proposing transferable skill clusters to three manufacturers within 60 days. It didn't go as planned; two manufacturers wanted welders, a skill only 40 of 320 workers held. Then came reflection: the gap wasn't skills, it was information — they had mapped supply before demand. Then consultation, bringing in the state, the college, and the HR directors who had declined. Out of that came a standing labor exchange, updated quarterly. The community didn't produce one clever solution. It produced a growing capacity to solve problems it hadn't yet encountered.

The Institution is the architect of empowerment. Its job shifts from command-and-control to platform-and-enable: directing resources toward expanding what individuals and communities can do rather than substituting institutional judgment for theirs. Mondragon — founded in 1956 with 24 workers, now roughly 70,000 across an integrated network with €11.2 billion in 2024 revenue — is institution-as-architecture, not institution-as-command.

Each cycle restarts at higher capacity than it began. Empowered individuals take initiative. Communities convert those initiatives into collective learning. Enabling institutions gather patterns across communities and create conditions for the next round. The direction is neither top-down nor bottom-up. It is lateral and iterative.

The most urgent gap: governing the machine that acts

Here the architecture meets its hardest unfinished assignment. Every national AI framework written before 2024 quietly assumes a system recommends and a human decides. That model still exists, and where it holds, the rules mostly work. But a different kind of AI now operates at commercial scale: autonomous agents that plan multi-step tasks, call APIs, execute transactions, and take consequential actions without human approval at each step.

When an agent autonomously denies a loan, the appeal pathway evaporates — there is no human decision-maker to appeal to, no auditable trail. The International AI Safety Report 2026, led by Yoshua Bengio with more than 96 researchers across 30 nations, names exactly this: a growing mismatch between the speed of capability and the pace of governance. This is not a gap in enforcement. It is a gap in conception. The rules were written for a machine that talks. They don't yet govern a machine that acts. Closing it is the Institution protagonist's most urgent work.

What to do with this

• Stop asking which level should act. Ask how the three coordinate. In Cleveland, the Evergreen Cooperatives emerged because a foundation identified the procurement gap, anchor institutions committed dollars before a single load of laundry, and displaced workers accepted equity instead of wages.
• Run the loop, not the plan. Act in small reversible steps; examine what the gap between plan and outcome reveals about your assumptions; then open the next decision to people who hold information you don't.
• Use Participatory Technological Assessment before deployment, not after harm. The people most affected by a technology carry knowledge no technical expert can fully substitute. Inviting them to interrogate an automation deployment produces governance that earns legitimacy rather than purchasing it.
• Name the agentic governance gap in your own systems. If an autonomous agent makes a consequential decision, where is the mandatory human checkpoint, the auditable liability, the external verification? Voluntary self-certification is not governance.

The principle

The dominant framing of AI discourse treats the technology as the subject and institutions as secondary constraints. The architecture of agency reverses the emphasis. Technology generates capacity; the architecture for distributing what that capacity produces is what determines whether displacement becomes despair or reorganization. Remove any one protagonist and the structure collapses inward. Individuals without community support drift. Communities without institutional guarantees protect only those who already have cushions. Institutions without individual formation produce compliance, not protagonists. The three are load-bearing — all of them, simultaneously.