Who Builds the Agents

Fifty-six million people use Microsoft Power Platform every month. Up 27 percent in a year. Most of them have never opened a terminal.

IDC’s research VP said it this summer: low-code and no-code platforms are now the primary way companies build and run AI agents at scale. Not a growing channel. The primary one.

I keep running into the same pattern with clients. Leadership assumes agent deployment is an engineering project. Custom integrations, dedicated team, months of runway. That assumption made sense two years ago. The tooling has caught up. The mental model hasn’t.

Navien saved 28,000 hours a year with agents built by employees who never called IT. Grupo Bimbo built two agents during a hackathon — audit time dropped 20 percent. Atos runs 19,000 agents through a unified operating model and cut incident triage by 68 percent. None of these started with an engineering hire.

The failures I see follow a consistent shape: technically sound agents that don’t work operationally, because the builder never did the job the agent was supposed to improve. The person who knows the process builds a better agent than the person who knows the stack. Every time.

Gartner says low-code spend hits $58 billion by 2029. The variable isn’t budget or model capability. It’s whether you trust the people doing the work to build the tools that change it.