Over the last ten years, most large organisations built a second workforce without ever deciding to. It isn’t made of people. It’s bots, RPA, scripts, APIs, models and now agents. Each one was bought to fix a real problem and most of them still do. Together they now run a growing share of the operation.
Your human workforce has managers, objectives, service levels and reviews, and someone accountable when it goes wrong. The digital one has none of that as a whole. It was put together a tool at a time, by different teams, solving different problems, and nobody manages it as one thing.
So the useful question isn’t how many agents you’re running. It’s who is accountable for managing them as a single workforce. In most organisations the honest answer is nobody.
It’s gaining pace
Gartner put the average large enterprise at fewer than 15 agents as recently as last year. IBM’s research already has this year’s figure past 1,600. Gartner expects Fortune 500 companies to be running more than 150,000 agents each by 2028.
Scale isn’t the whole story. Roughly six in ten organisations report evidence, or at least a strong suspicion, that agents are running inside their business without proper authorisation. Nobody signed them off and nobody is watching them. IDC research suggests close to nine in ten agentic pilots never reach production at all.
Most of you are carrying two problems at once. There’s a mature RPA estate under pressure to justify its cost, and a new wave of agentic work with no operating model underneath it. Every new technology becomes one more silo, treated as a science project when it should be production infrastructure.
Weaving both together under one independent layer of control, insight and governance is the actual challenge for most enterprise automation teams right now.
Why the referee can’t be on the player’s payroll
Every major platform is moving up the stack into orchestration and AI, building a control layer on top of its own execution engine. The pitch is convenience. The result is that you’ve moved the lock-in one level higher and called it progress.
Think about what you’re asking a vendor to do when orchestration sits inside their own execution platform. You’re asking them to mark their own homework. You want them to tell you honestly when their product is the wrong choice, when a competitor would save you money, or when you should buy less of what they sell. That’s an unfair thing to ask of anyone. It has nothing to do with the integrity of the people involved, most of whom are good and honest. It’s structural. Put the referee and the player on the same payroll and objectivity is gone before the whistle.
Independence isn’t a feature you add. It’s the reason the layer exists.
The analysts got there on their own
A year ago we said orchestration couldn’t keep living inside execution platforms without becoming another form of lock-in. At the time that was one vendor’s opinion. Since then the market has started saying it too.
Forrester splits enterprise agent architecture into three planes: build, orchestrate and control. Their view is that the control plane is still underdeveloped and immature, and that no vendor yet delivers it as a genuinely independent platform. They also found that most vendors already support open standards like MCP and A2A, and they read that as a reason for independent governance to matter more. The more interoperable the ecosystem becomes, the more valuable it is to have something above it that sees across all of it. Governance built inside a single platform stops at that platform’s walls.
Gartner expects 60 percent of organisations to run orchestration on a platform from a vendor outside their existing systems of record by 2030. They’ve also named the category: Agent Management Platform. They rate the benefit of a universal orchestrator as transformational and the market maturity as embryonic, with penetration somewhere between 5 and 20 percent. They’re also clear that no single provider delivers a comprehensive solution yet.
We read this as white space, and we’ve been standing in it for a while.
What a real control layer has to do
If the layer that proves your ROI can’t be gamed by the vendor selling you the technology, three things have to be true.
Orchestration comes first, because it’s the one most people think they already have. Take any process in your business. A bot picks up the request, an agent reads a document and decides what happens next, a script moves data between two systems that were never designed to talk, an API fires, and a person makes a judgement call. Today each of those steps is usually managed separately, if at all. Everybody owns their piece and nobody owns the process. Real orchestration is one layer that runs the whole thing end to end, whatever technology is doing the work at that moment. Many of our customers run this through universal queues, one queue model across the estate instead of one per platform, which is why some of them no longer keep a fallback system.
Operations keeps the work honest, compliant and secure while it’s running, in real time rather than in a report someone builds at month end. It means security and governance enforced consistently across every technology in your estate, including the ones no single vendor sells you. One of our customers put it better than we could: “it’s the one percent that hurts.” It’s the bots that miss a deadline and the agents that burn through a month’s token budget in an afternoon. Operations exists for that one percent, and the discipline is the same whether the thing that failed was built ten years ago or ten weeks ago.
Intelligence proves what all of this costs and what it delivers, in a number your board will trust. You’re already paying for automation in currencies that don’t compare. There are licence seats, infrastructure, people’s time and now tokens, which don’t even compare with themselves, whether input against output, cached against fresh, or an agent against the same task done the old way. Intelligence turns that into one number per process, per team, per pound spent, that means the same thing wherever it came from. It’s not an estimate and it’s not a vendor’s dashboard telling you their own product is worth it.
One question to take back to your team
Stop asking which platform to add next. Ask who owns your control layer, and whether that’s who you chose or who you ended up with by default.
At C TWO, we believe the control layer has to be independent, so we built it.
On 15 October we launched the universal orchestration layer: orchestration, operations and intelligence across every bot, script, API and agent in your estate, with measurement outside any single vendor’s dashboard.
Over 170 customers across 14 countries already run this way.
