Citizen Agents and Flagship Agents: The Portfolio You Have Not Counted
Last week I closed a keynote in Melbourne with a slide that asked one question. How many AI agents are running in your enterprise right now? Before I could answer it, a senior CIO at one of the tables said, loud enough for the front row to hear, "we have been living in it."
I have not been able to put that line down since. It has been organising every conversation I have had since I came off stage. Most of those conversations end at the same problem, which is one I had not seen clearly before. I want to walk you through it this week, because I want to know whether you are seeing it too.
Almost every CIO I talk to is dealing with two AI programmes at once, and almost none of them are seeing both. One programme they commissioned. One they did not. Both are in production. Both are drifting. The governance models I see in most enterprises are built for one of the two and blind to the other.
The names I have been using lately are citizen agents and flagship agents, and I want to test them with you here.
The Failure Anatomy
The first one came from a compliance officer at a wealth management firm. We had been corresponding for weeks about something else when she wrote one evening with a line I read three times: "we have three different versions of our investment advice running across the same clients, and we did not know until audit asked."
Her firm had rolled out Copilot Studio across their relationship managers eight months ago. Each RM was free to build their own client briefing agent. Pull from the CRM, summarise portfolio performance, prep talking points for the next meeting. It worked. RMs loved it. Productivity numbers looked good. Then audit happened. Three different RMs had built three different versions of the briefing agent with different prompts, different data sources, and different definitions of high concentration risk. All three had been used to brief clients across the same families of accounts during the same quarter.
She did not know how many other agents her RMs had built. Nobody did. No registry, no risk classification, no cost tracking. Just a platform, a few hundred enabled users, and an audit finding that exposed eight months of ungoverned advice.
This is the bottom-up wave. Hundreds of small agents built by individual employees on platforms like Copilot Studio, Google AgentSpace, Salesforce Agentforce and ChatGPT Enterprise. Each agent on its own is small. The aggregate portfolio is enormous and largely invisible. These are citizen agents.
The second came from a credit risk lead at a bank. We were on a call about something else when he said, almost in passing, "we own the model on paper, but I cannot tell you who actually runs it." I made him repeat it because I knew immediately I would be quoting it.
His bank had a credit decisioning agent supporting the underwriting team. On paper his function owned it. The model registry showed credit risk as the accountable team. In practice, performance monitoring sat with the data team, cost monitoring sat with finance, and compliance review sat with the second line of defence. When drift showed up, none of those three functions had the full picture, and his team did not have the technical depth to act on what they could see. The agent was visibly drifting and structurally unowned.
This is the top-down wave. A handful of high-stakes, multi-team, customer-facing or revenue-critical agents commissioned through formal programmes. Each agent is large in scope. The portfolio is small in count, strategic in profile, and visible to the executive committee. These are flagship agents.
The Framework Lens
Here is what I keep coming back to. These two waves are happening inside the same enterprises at the same time, and the governance instincts each of them triggers are pointing in the wrong direction.
When the citizen wave arrives, IT and procurement see a tool sprawl problem. The instinct is to lock down the platform, which is the right reflex for shadow IT but the wrong frame for citizen agents, because citizen agents are decisions automated by individuals, running across customer interactions and internal workflows, drifting and recompiling without anyone watching. You can lock down the platform and still have a thousand agents in production making different choices about the same business problem.
When the flagship wave arrives, the executive committee sees a programme delivery problem. The instinct is to assign clear ownership and a steering committee, which is the right reflex for legacy software but the wrong frame for flagship agents, because flagship agents fail through silent drift, distributed accountability, and trust erosion. The steering committee meets quarterly. The agent drifts continuously. The two are operating on different clocks.
The frame I keep arriving at is that these are two halves of the same portfolio, and the discipline that needs to govern them has to operate on both at once.
A portfolio is a portfolio. Until the enterprise sees it that way, the discipline cannot form.
The Field Test
If you want to see this for yourself in your own organisation, the cleanest way I have found is one question.
Walk into your next executive committee or AI governance forum and ask: how many AI agents do we have in production right now? I want both numbers. The flagship count and the citizen count.
Here is what tends to happen. The room goes quiet. Either nobody knows the citizen count, or somebody starts to estimate, or somebody insists there are no citizen agents because the platform is locked down. I have seen all three answers in the last six weeks. They all tell you the same thing, which is that your enterprise is running a portfolio it has not counted.
A working operating model produces both numbers without hesitation. One name accountable for each flagship agent. One named function counting every citizen agent against the same standard. If neither number exists, the portfolio is running the enterprise, not the other way round.
The Signal
Reply with one line. Tell me whether your enterprise has a citizen count, a flagship count, or neither. I will anonymise. The aggregate becomes the first practitioner-grade view of how the two waves are actually distributed in 2026, and where the governance is failing first. This data does not exist anywhere yet.
The full case for governing both waves runs across The Pilot Trap, where I built out the operating manual: how flagship agents fail, how citizen agents accumulate, and what discipline holds them together inside one portfolio. (https://www.amazon.com/dp/B0DF5YVR4H)
The discipline behind this is forming into a practitioner working group later in 2026. A small first cohort of CIOs, CAIOs, and AI delivery leaders working through the operating model together, contributing to the field as it takes shape. If you want to be in the first cohort, reply with one line. The criterion is whether you are inside the problem and want to shape the answer.
One thing I am still working out. The line between citizen agents that need governance and citizen agents that should just be allowed to die without intervention is not clean. Some of them are productivity hacks that should never be touched. Some of them are eight months of ungoverned investment advice. The discipline I am building does not yet draw that line precisely, and I think it is the next problem to solve. If your organisation has a way of telling them apart that works, I want to hear it.
Next issue. There was a second question that kept circling the room that day. Why do AI pilots succeed and production deployments fail? The structural answer is what Issue 7 is about. Before pilot and production sits a layer almost no enterprise has built. The Control Tower.
Vijayan Seenisamy
Enterprise Agentic AI Systems Delivery | Creator, AI ROF (TM)
Author, The AI Delivery Manager Blueprint and The Pilot Trap
New issues arrive by email first. Subscribe here.
