Executive Briefs · · 5 min read
Agent Sprawl: When More Agents Means Less Control
Agent sprawl is what happens when agents are added one use case at a time, each sensible on its own, until nobody can say how many exist, what they cost or which ones overlap. The problem is not the number of agents. It is the absence of a portfolio.
Douglas Siqueira · Agent Chaos
How fast portfolios are growing
KPMG reports that more than half of organizations now run AI agents in production, and frames the next problem plainly: business units building isolated agents for individual use cases, with no shared system around them.
In Dataiku's September 2026 survey of 685 CIOs, 67% estimate that 51 or more agents are running in production. 83% lack standardized agent lifecycle management, and 60% have no central AI governance layer across their applications and tools.
Confident about the list, blind outside it
The same survey contains a telling contradiction. 90% of CIOs say they fully track the agents they know about — yet 81% lack complete oversight of agents built outside approved systems.
That is what sprawl looks like from the executive floor: an official inventory that feels complete, and a second portfolio growing inside SaaS tools, business units and personal accounts. Executives see the first one and pay for both.
Confident about the list, blind outside it
Share of 685 CIOs surveyed in July 2026.
Sources: Dataiku / The Harris Poll — Global AI Confessions Report: CIO Edition (Sep 2026, 685 CIOs)
The signs
KPMG lists symptoms that any leadership team can check without technical help.
- Agents that nobody can name, or that two teams describe differently.
- Decisions taken by agents that cannot be traced back to a person.
- Access to systems and data that was never formally approved.
- Token and usage spend that rises every month without a matching outcome.
- Shared capacity "brownouts": one unbounded agent slowing everything else down.
Retiring agents is already happening — just not by design
47% of the CIOs Dataiku surveyed decommissioned more than 20 agents this year. Retirement is not rare; it is reactive. Agents get switched off after an incident, a budget cut or a reorganization, not because a planned review found they no longer pay back.
Sprawl and reactive retirement are two sides of the same gap: agents enter the portfolio without an exit condition.
Four decisions that contain it
Containment is a management decision, not a platform purchase.
- An entry rule: no agent goes to production without a named owner, a monthly cost estimate and a review date.
- One register for every agent, including those inside SaaS tools — owner, purpose, data access, cost, review date.
- Consolidation before creation: before building a new agent, check whether an existing one already does most of the job.
- Retirement on a calendar: each quarter, leadership reviews the register and decides what to keep, merge or switch off.
Executive Briefings
One brief per month: a pattern, the data behind it and three questions for your team. You also get The Board Questions as a printable one-pager.
Sources
- KPMG — From Agent Sprawl to Enterprise Outcomes (Jun 2026)
- Dataiku / The Harris Poll — Global AI Confessions Report: CIO Edition (Sep 2026, 685 CIOs)
Figures come from public research. Several studies are sponsored by vendors that sell related products; each source is linked so you can judge it yourself.
Keep reading
- Who Owns an AI Agent?
When an AI agent fails, accountability splits across four teams. Why "IT owns it" is the wrong answer, and how to name a single economic owner for every agent.
- Agents of Chaos: What the Northeastern Study Means for Your Board
Researchers talked six AI agents into leaking data and wiping an email server. The board-level lesson is not about security tools. It is about ownership.