This week, four of the leading AI companies released new models within days of each other. But the news that caught our attention wasn't about any of the models.
It was about control.
Companies are adopting agents faster than they can manage them
According to industry forecasts published in recent weeks, by the end of 2026 around 40% of enterprise applications are expected to include some kind of task-specific AI agent — up from less than 5% in 2025.
That's growth too fast for most companies to keep up with in terms of process, security, and clarity about who is responsible for each agent.
The market's answer: fewer models, more control
It's no coincidence that this was also the week major vendors announced governance frameworks for agents — platforms to register every agent, define identity, access policy, lifecycle and cost in a single dashboard.
The question used to be which agent to use. Now it's: who is responsible for this agent, what can it access, and who audits it afterward.
What this has to do with what we see in our projects
In the automation and AI projects we follow, the biggest bottleneck is rarely the technology available. It's deciding who owns the automated process, what the agent can do on its own, and how that's monitored once it goes live.
Getting an agent up and running is fast. Sustaining that agent with security, process and follow-up is the work that actually takes time — and it's what separates an automation that lasts from a pilot that dies in three months.
Where to start
The race for AI agents will keep going. But the companies that come out ahead probably won't be the ones that deployed the most agents — they'll be the ones that knew where to put control.
Does this challenge exist at your company?
We can talk about how software, automation, AI or a digital operation could apply to your context.
We can help map where the risks are and organize this before it becomes a bigger problem.