Agent Sprawl: Enterprises Repeating Silo Mistakes, Gartner's Governance Fix

Enterprises are building AI agents in the same fragmented way they built isolated systems two decades ago: finance, HR, customer service, and R&D each spin up their own agents. While each agent delivers local value, the aggregate creates Agent Sprawl — a term coined by Gartner — leading to duplicated functions, tangled data access, unclear ownership, and governance gaps.

From System Silos to Agent Silos

McKinsey's research on Agentic Organization shows a 2–5 person team can already oversee 50–100 specialized agents running end-to-end processes like customer onboarding, product launches, and financial close. Yet McKinsey warns:

"Proliferation of AI agents … can be a recipe for chaos."

When agents number in the dozens or hundreds, questions multiply: Why does this agent exist? Does it duplicate another? What data can it read or write? Which systems and APIs does it call? Where does its knowledge come from? Are skills reusable? Who validates model upgrades? Who is accountable for failures?

McKinsey further notes (September 2026):

"Many enterprises are scaling AI agents faster than they are redesigning the work underneath."

This mirrors the earlier IT era where every reasonable project eventually produced system and data islands. Today the risk is Agent islands, Knowledge islands, Skill islands, wrapped in models, APIs, permissions, and workflows.

Gartner (April 2026) predicts that by 2028 the average Fortune 500 company will run over 150,000 agents , up from fewer than 15 in 2025. The challenge shifts from "do we have agents?" to "how do we govern them at scale?"

Gartner's Six Governance Steps

Gartner prescribes six actions to balance control with innovation:

Establish agent governance rules and policies

Create a centralized agent registry

Manage agent identity, permissions, and lifecycle

Strengthen AI information governance

Continuously monitor and correct agent behavior

Build a responsible AI culture

Gartner summarizes:

"Organizations need to find a balance where they can govern agents and manage sprawl, but also safely empower employees to innovate."

Three Relationship Categories for Enterprise Architecture

The author maps Gartner's steps into three architectural relationship layers:

1. Business Relationships

Every agent must answer: What business capability does it serve? Which process step? What problem does it solve? Who performed this before? Without clear answers, agents become "AI for AI's sake."

2. Capability Relationships

What data, knowledge, applications, APIs, tools, and models does the agent depend on? How does it relate to other agents? This layer determines true reuse; otherwise each department rebuilds its own knowledge bases, skills, and integrations.

3. Governance Relationships

Who owns the agent? What permissions (read/write) does it hold? Who monitors it? Who is accountable for incidents? When should it be merged, adjusted, or retired? This embeds Gartner's registry, identity, lifecycle, information governance, and monitoring into the overall enterprise architecture.

The core insight: AI-era enterprise architecture does not need a new "AI layer" — it needs to place every agent back into the existing architecture so each can answer: Why exist? What business? What capabilities? What permissions? Who owns? If these relationships are managed, 100 or 1,000 agents are merely a scale problem; if not, even a few dozen create new islands. The lesson from 20 years ago: don't wait until agents are built to start governance.

Reference:

https://www.gartner.com/en/newsroom/press-releases/2026-04-28-gartner-identifies-six-steps-to-manage-artificial-intelligence-agent-sprawl