The Agentic Operating Model - Eleven Realities

Every software company wants to push its teams toward agentic AI, hoping to get faster. Ship more features in less time.

In this series, "The Agentic Operating Model", I want to show you what that transformation actually looks like. Where the value is. Which pitfalls nobody warns you about. And how to make it work in your organisation.

Let me start with what I've been hearing on the road as a consultant over the past few months. All of these are real. I didn't invent a single one.

• Company owner: "I built our whole company app over the weekend. Prompted it in Lovable. Want to see it? Why are these development teams so damn slow?"
• CEO: "We don't need requirements anymore. With my domain knowledge I can prompt them myself, and a few developers can generate the whole application in days."
• CTO: "Finally I can build things myself. Patrick, I'm going back to the engine room."
• CFO: "Finally we can cut IT cost to the bone."
• Product owner: "These developers don't understand the domain. Communication is a hassle. I'd rather just work with agents."
• Requirements engineer: "Finally I can reverse engineer the entire codebase and have Claude show me the drift - everything the developers didn't build the way the requirements said."
• Software architect: "We have to invest heavily in the harness around agentic AI. Otherwise this will fail."
• Senior developer A: "For repetitive work it makes me faster. For small things I'm quicker on my own."
• Senior developer B: "Finally I don't have to document anything anymore. Claude can do it. Yes!"
• Senior developer C: "This trend scares me. I'm looking for another job."
• Junior developer: - already let go. The AI-first transformation concluded there's no need for juniors anymore.

That's the actual story here: this isn't a tooling change.

Eleven people. Eleven completely different realities.

I stopped reading these as single opinions. Grouped together, they say something none of them says alone.

Three of them saw a demo and concluded product development is fast now.

The owner. The CEO. The CFO.

They're not wrong about what they saw. A working app over a weekend is real. Two years ago it wasn't possible.

But the prototype was never the expensive part. The expensive part is year two. Edge cases. Data migration. On-call. The auth model that has to survive an audit.

Agentic AI made the cheapest phase cheaper. It has barely touched the expensive one.

Five of them are celebrating that they don't have to talk to a colleague anymore.

Read them again.

The CEO doesn't need requirements engineers. The PO doesn't want to deal with developers. The RE finally has a way to catch developers. The CTO is going back to the engine room. The developer never has to write for a reader again.

That has nothing to do with AI. These people were frustrated by coordination long before Claude existed.

Agentic AI just gave everyone a plausible way out of a conversation they never wanted to have.

And notice: each of them wants to route around a different colleague. Not one names themselves.

Two of them talk about cost instead of speed.

The architect, who wants to build the harness first. And the senior developer who actually measured it - faster for repetitive work, slower for small things.

These are the only two answers in the list with a condition attached.

And then there's the eleventh.

The junior developer isn't quoted. There was no one left to quote.

That role got cut first, and the argument was easy: agents produce junior-level work for a fraction of the cost.

Here's what the argument misses. Reviewing agent output is the skill this whole transformation depends on. It's a senior skill. And every senior in this list got it by being a junior somewhere for a few years.

We're cutting the training pipeline for the exact skill we're about to need most.

So that's the tension.

In this series, "The Agentic Operating Model", I want to show you what that transformation actually looks like. Where the value is. Which pitfalls nobody warns you about. And how to make it work in your organisation.

Not as a tooling rollout. As an operating model. That's the part most companies get wrong, and it's where I spend my days.