Working Title: AI Agent Sprawl: By 2028, Your Enterprise Will Have 150,000 Agents. Is It Ready?
—
In 2025, the average large enterprise had fewer than 15 AI agents. Most were experimental — a customer support chatbot here, an internal search assistant there.
By 2028, Gartner predicts, that same enterprise will have over 150,000 agents.
Read that again. 15 to 150,000. A 10,000x increase in three years.
If you think managing 50 SaaS applications was a governance challenge, wait until you’re managing 150,000 autonomous AI agents.
What Is Agent Sprawl?
Agent sprawl is the uncontrolled proliferation of AI agents across an enterprise. It manifests in several ways:
Shadow agents. Employees build and deploy agents without IT knowledge or approval. They connect to company data, make autonomous decisions, and operate outside any governance framework. Sound familiar? It’s shadow IT, but with decision-making capability.
Duplicate agents. Three different teams build agents that do essentially the same thing — extract data from PDFs, monitor system health, generate weekly reports. Each was built independently, each has different quality standards, and each costs money to run.
Zombie agents. Agents that were built for a specific purpose, served that purpose, and are still running — even though nobody remembers why. They consume tokens, access data, and occasionally produce outputs that nobody reads.
Conflicting agents. Two agents with overlapping domains that give contradictory answers to the same question. Which one does the user trust? Neither, usually.
The Math of 150,000 Agents
Let’s make this concrete. A Fortune 500 company with 150,000 agents:
- ~300 agents per business unit (assuming 500 business units/teams)
- At an average cost of $50/month per agent (conservative): $7.5 million/month in agent runtime costs
- Each agent accessing an average of 3 data sources: 450,000 active data connections to govern
- Each agent making ~100 decisions per day: 15 million autonomous decisions per day
Without governance, this is chaos. With governance, it’s a superpower.
Gartner’s 6 Steps to Manage Agent Sprawl
Gartner’s April 2026 research identifies six critical steps:
1. Discover
You can’t govern what you can’t see. The first step is a comprehensive audit: what agents exist, where they run, what data they access, and who owns them.
Reality check: Most enterprises discover 2-3x more agents than they expected.
2. Inventory
Create a centralized registry of all agents. Each entry should include: owner, purpose, data access, cost, version, and status (active/retired/pending review).
3. Classify
Not all agents are equal. Classify by:
- Criticality (business-critical vs. nice-to-have)
- Autonomy level (assistive vs. autonomous)
- Data sensitivity (public vs. confidential data access)
- Cost tier (high/monthly spend vs. low)
4. Govern
Establish policies for agent lifecycle:
- Onboarding: New agents must be registered, reviewed, and approved before production
- Monitoring: All active agents must have quality metrics and cost tracking
- Retirement: Agents without an active owner or purpose are retired within 30 days
- Security: All agents must pass security review proportional to their data access level
5. Monitor
Continuous monitoring of agent health, quality, and cost. Aggregate dashboards showing total agent count, total spend, and quality metrics across all agents.
6. Optimize
Consolidate duplicate agents. Retire zombies. Optimize costs by sharing infrastructure across agents. Invest in an „agent platform“ rather than point solutions.
The Agentlake Concept
Nutanix has proposed the concept of „Agentlakes“ — a centralized data layer that all agents draw from, rather than each agent having its own data silo. The benefits:
- Single source of truth: All agents access the same data, reducing inconsistencies
- Simplified governance: Control data access in one place, not across 150,000 agents
- Cost efficiency: Shared data infrastructure is cheaper than 150,000 separate connections
- Auditability: All data access is logged in one place
First-Mover Patterns
Enterprises that are ahead on agent governance share common patterns:
Agent Registry: A single source of truth for all agents, their owners, and their status. Some teams use a simple spreadsheet; others build dedicated tools.
Agent Procurement Workflow: A lightweight process for requesting, approving, and deploying new agents. Like a software procurement process, but faster.
Agent Lifecycle Policies: Clear rules for when agents are created, reviewed, updated, and retired. No agent runs without an owner.
Agent Platform Team: A small team responsible for the shared infrastructure, tooling, and governance framework that all agents build on.
Quick Win: Start with an Inventory
If you do nothing else this quarter, do this:
1. Survey every team: „Do you have any AI agents in production?“
2. For each agent, record: name, owner, purpose, data access, monthly cost
3. Identify duplicates, zombies, and shadow agents
4. Present findings to leadership with a governance proposal
You’ll be surprised what you find. And you’ll be building the foundation for governing 150,000 agents before you have 150,000 agents.
The Bottom Line
Agent sprawl isn’t a future problem. It’s happening now, as enterprises rapidly deploy AI agents without the governance frameworks to manage them. The enterprises that invest in agent governance today — discovery, inventory, classification, governance, monitoring, optimization — will be the ones that can scale to 150,000 agents without descending into chaos.
The ones that wait will spend the late 2020s cleaning up the mess.
—
Word count: ~1,050 (excerpt — full draft would expand with more case studies, governance framework details, and implementation roadmaps to reach 1,800-2,200 words)