Blog The Unexpected Challenge of AI Success: What do you do with 9,000 agents?

As a Client Zero for Microsoft’s AI technologies, one of Insight’s responsibilities is learning what happens after adoption. We’re using our own environment as a proving ground, testing new capabilities, uncovering challenges, and developing repeatable approaches before bringing them to clients.

For the last few years, most conversations around AI have focused on how to get people to use it.

At Insight, we had the same conversations, so we developed the AI Flight Academy to turn employees into regular, responsible, AI users. We gamified the learning experience and made AI a workplace tool everyone could access.

We encouraged experimentation. Employees responded by creating AI agents to assist them with their work. Now we needed visibility into what had been created, how it was being used, and where governance needed to evolve.

This wasn’t something I set out to investigate because of a security incident or compliance concern. It started because our success with AI adoption created a new set of questions. As our agent count climbed into the thousands, I started wondering: What exactly exists in our environment?

When I first started looking, we had thousands of agents. By the time I ran my latest analysis, we were approaching 9,000.

People create agents to solve problems in their own workflow. Teams build their own automations. Departments experiment, refine, and improve what works for them. This isn’t shadow AI. This is what we’ve enabled teams to do — and now they’ve done it, and we need to figure out what we’re going to do with the results.

Every new agent introduces governance questions:

• Does this capability already exist somewhere else?
• Who owns it?
• Which data can it access?
• Is it being used?
• Should it be promoted, consolidated, or retired?

We needed the same lifecycle discipline for AI agents that we apply to applications, infrastructure, and data assets. The problem was that we didn’t have a way to apply that discipline at scale.

Visibility comes first.

As a security practitioner, I’ve always believed a simple principle: Governance is impossible without visibility. Or, as I sometimes say, “I don’t know what I don’t know. You can’t protect what you don’t know about.”

That’s why Microsoft Agent 365 had already become a foundational part of our AI governance journey. As part of our broader E7 and Client Zero efforts, we were using Agent 365 to understand our rapidly expanding AI environment, discover agents across the organisation, and gain visibility into how those agents interacted with data, tools, and users.

More importantly, Agent 365 gave us a way to see beyond the agent itself. We could understand the instructions driving it, the knowledge sources informing it, the tools it could access, the connections it relied on, and how it was actually being used.

But as our adoption accelerated, I started seeing another opportunity. If Agent 365 could show us what existed, maybe it could also help us understand what was valuable, what was redundant, and what deserved more attention.

Agent 365 gave us the ability to see our AI estate in a way that wasn’t possible before. Beyond a list of agents, it provided a look into the details I needed, including:

• Agent instructions
• Knowledge sources
• Connected tools
• Data connections
• Ownership information
• Usage activity

For the first time, we had the visibility to treat agents as enterprise assets rather than isolated experiments. That changed the conversation from, “How many agents do we have?” to “Which of these agents are valuable, might introduce risk, or deserve broader attention?”

What we discovered

The results were Client Zero gold. The governance challenges we uncovered aren’t unique to Insight. They’re the challenges any organisation will face as agent adoption matures.

Some findings were exactly what you’d expect when nearly 9,000 agents exist across a large organisation. We found thousands of agents that appeared to have little or no activity; multiple agents solving nearly identical problems; and clusters of RFP, proposal, and SOW agents that had been built independently by different teams.

In many cases, employees were solving the same challenge repeatedly because they didn’t know a similar solution already existed.

But the most interesting discoveries weren’t the redundant and abandoned agents. In several cases, we found agents with only one single active user but thousands of sessions — which means someone, somewhere in the organisation, has quietly built something extremely valuable. They’ve integrated it into their daily workflow, they’re using it constantly, and nobody knows it exists. That’s an innovation opportunity.

One of the biggest misconceptions about governance is that it’s only about finding risk, but good governance also helps organisations identify what’s working and scale it.

The next phase of AI maturity

What surprised me wasn’t that people built thousands of agents. That’s exactly what we hoped would happen. What surprised me was how quickly the challenge shifted from encouraging innovation to managing it.

The next phase of AI maturity isn’t about creating more agents. It’s about understanding, governing, and continuously improving the ones you already have.

Organisations already do this with applications, infrastructure, and data assets. AI needs the same discipline.

The lifecycle looks like this:

Create → Discover → Classify → Govern → Optimise → Retire

• Know what exists.
• Understand its purpose.
• Identify ownership.
• Determine whether it’s creating value.
• Make decisions about its future.

Without that process, agent ecosystems naturally become larger, more fragmented, and harder to govern over time.

Why this matters for every organisation

Many organisations are heading toward the same destination. Today, a business that has dozens of agents might soon have hundreds. Eventually it may have thousands.

As our own AI ecosystem continues to grow, that’s the challenge we’re solving as Client Zero. We’re developing the governance practices, visibility models, and lifecycle processes that we’ll ultimately help our clients implement as their own agent ecosystems expand.

The organisations that succeed in the next phase of AI maturity won’t necessarily be the ones with the most agents, they’ll be the ones that understand:

• Which agents matter
• Who owns them
• What they can access
• Where they create value
• When they should evolve and when they should be retired

Ready for the next phase of your AI journey?

The challenge facing many organisations is no longer whether employees will adopt AI. It’s how to maintain visibility, governance, and control once adoption succeeds.

Microsoft Agent 365 helps organisations establish that visibility by providing insight into agent inventory, ownership, instructions, knowledge sources, usage, and connections. Combined with Insight’s Client Zero experience and governance expertise, organisations can move from experimentation to sustainable AI operations.