Gartner forecasts that by 2028 an average Fortune 500 enterprise will have more than 150,000 AI agents in use. Only 13% of organisations believe they have the right governance in place (Gartner, 2026). The gap between those two figures is a question of AI agent governance capacity. An organisation can safely run only as many agents as its accountable humans can govern, whatever its deployment budget allows.
In Brief
- A board report that counts only deployed agents hides how many of them no accountable person governs.
- Accountable humans, each with authority and current instructions, cap the number of AI agents that can run safely.
- Committees and policy frameworks above ungoverned agents add no capacity; each level of the chain needs a single, accountable owner.
- Capacity grows through better skill files and a tighter Definition of Done, or through new value streams with their own Product Owners.
The question executives ask is: how many agents can we deploy? The question that determines whether those deployments hold is different: how many agents can our accountable humans actually govern? Agent deployment grows with investment, while governance capacity grows with the number of people whose judgment the organisation can count on when something goes wrong.
Accountability creates the ceiling
The number of agents an organisation can safely run is set by the number its accountable humans can govern (Hodgson & West, 2026). To govern, each of them needs defined authority, regular inspection of the agents’ work and governing instructions kept current.
Every agent beyond the governance capacity ceiling is ungoverned by definition: the organisation has deployed AI that no accountable person can answer for.
The accountability chain runs from the Developer governing an agent team to the portfolio executive accountable for investment across products (Hodgson, 2026). Between those levels sits the Product Owner, accountable for value and release decisions. Governance capacity depends on that chain holding at every level. A central AI governance committee, a policy framework or an audit structure placed above ungoverned agents does not substitute for it. Each level needs a single, accountable owner, not a committee, working group or team, with the authority and the governing instructions to act.
Spending on AI infrastructure raises the agent count without raising the number of accountable humans, and each person’s judgment covers only so many agent teams. Every approval past that point adds output to govern and nobody to govern it. The limit sits in the accountability structure, where a larger technology budget does not reach.
From inside a planning cycle, the gap between deployment and governance does not surface as a governance problem. The organisation instead discovers it cannot explain the output: agents producing results nobody fully understands, and decisions reaching the board before any accountable person reviewed them. Audit trails satisfy the form of accountability without its substance. By the time the board sees those output failures, the ungoverned agent count has grown past what any retrospective fix can address.
Every ungoverned agent is a liability
A board report on AI agents needs two counts: agents approved, and agents covered by an accountable person with the authority to act. The difference between those counts is the ungoverned volume. A report that shows only the deployment count leaves that volume out of the board’s view.
Every agent approved without a corresponding accountable person carries a liability that builds with each Sprint the gap stays open. The liability comes from the missing accountable structure: a single, accountable person with authority and governing instructions maintained in practice. When that structure is absent, the governance audit that follows will find agents in production with no one accountable for their output.
Governance capacity is the growth lever
An organisation can increase governance capacity through two routes without deploying beyond what its accountable humans can cover.
Route #1: Raise the governable ratio
As skill files (the written instructions agents work from) mature and the Definition of Done tightens, each accountable person can safely govern more agent teams (Hodgson & West, 2026). More precise governing instructions reduce the burden on each accountable person, and governance capacity grows with them. How large an agent team a Developer can govern depends on skill file maturity, the precision of the Definition of Done, and the complexity of the work. No evidence yet supports a specific ratio, so teams discover it Sprint by Sprint.
Route #2: Add governed value streams
Add products and value streams, each with its own accountable Product Owner and its own backlog, rather than adding oversight layers above ungoverned agents (Hodgson & West, 2026). Each new value stream comes with its own Developers and its own Product Owner, and agent volume follows.
Both routes grow agent capacity only as fast as accountable humans can absorb it, and neither deploys first and governs later. A board paper that reports agents governed beside agents deployed turns the choice between those routes into a funding decision, made before the ungoverned count grows.
References
- Gartner. (2026, April 28). Gartner identifies six steps to manage AI agent sprawl. Gartner, Inc. https://www.gartner.com/en/newsroom/press-releases/2026-04-28-gartner-identifies-six-steps-to-manage-artificial-intelligence-agent-sprawl
- Hodgson, M. (2026). Evolve: The operating model AI demands. Zen Ex Machina. https://zenexmachina.com/evolve/
- Hodgson, M., & West, D. (2026). The Developer’s Nexus: What happens to Scrum when every team member governs agents (Empiricism at Machine Speed, Paper 5). Scrum.org. https://www.scrum.org/resources/empiricism-machine-speed-5-part-series
