On Friday, July 17, 2026, I delivered a webinar on the Microsoft Reactor YouTube channel called Agentic DevOps Meets Accessibility: Inclusive Delivery with AI & Agents. The recording is now available, so you can watch it anytime.
Why the next chapter of FinOps is not about finding a cheaper model. It is about engineering systems that do not waste the tokens they already have.
A finance leader opens the monthly invoice for the company's AI platform and finds a number that does not match any story anyone can tell. Usage grew modestly. The bill grew sharply. Nobody switched to a pricier model. Nobody approved a new integration that anyone remembers. The line item simply grew on its own, the way cloud bills used to grow before anyone built a discipline around watching them.
Ask the engineering team what happened and the answer is rarely a single cause. It is a hundred small decisions: a system prompt that grew every time someone patched in a new rule, a retrieval step that fetches ten documents when two would do, an agent that retries a failing tool call five times before giving up, a workflow that hands a conversation between three specialized agents and resends the full history at every handoff. None of these decisions looked expensive in isolation. Together, they are the bill.
For decades, writing code was the expensive part. Reading it was almost free by comparison. A developer spent hours producing a change, and a reviewer spent minutes confirming it. That ratio shaped everything: our tools, our processes, our sense of who was busy and who was waiting.
Agents inverted that ratio. A capable agent now produces a complete branch with code, tests, and documentation in the time it takes to write a thoughtful task description. Authoring became cheap. Reading did not.
On Friday, June 12, 2026, I delivered a webinar on the Microsoft Reactor YouTube channel called DevOps Foundations for the AI Agentic Era. The recording is now available, so you can watch it anytime.
DevOps has always been defined by a simple, powerful equation: People + Process + Tools. That formula captured something essential about how modern software gets built and delivered. It broke down walls between development and operations. It gave organizations a mental model for diagnosing what was wrong when things moved too slowly, failed too often, or created too much friction.
For over a decade, this three-pillar model served the industry well. And it did so because it rested on an assumption that nobody questioned: every participant in the software delivery lifecycle was human.
That assumption no longer holds.
Not so long ago, every commit in your repository came from a human. A developer wrote code, pushed a branch, opened a pull request, and a reviewer approved it. Your CI/CD pipeline was designed around that flow: run tests, check lint, scan for vulnerabilities, deploy if green.
That assumption is breaking.
Most AI tutorials start with "create an Azure resource" and end with "here's your chat completion." They skip the messy middle — the part where a developer goes from "I wonder which model would work for this" to "this is running in production, monitored, secured, and costing what I expected."
That full journey is what this post is about.
AI coding agents are no longer experimental. Teams are using GitHub Copilot, Claude Code, Cursor, and other tools to generate code, open pull requests, review changes, and automate multi-step engineering tasks. The results are impressive, but a quieter problem is growing underneath the productivity gains.
Every developer on the team configures their AI agents differently.
AI-assisted development is no longer an experiment. It is part of everyday engineering work across organizations of all sizes. Tools powered by large language models generate code, propose refactors, write tests, summarize pull requests, and coordinate multi-step engineering tasks.
This creates a fascinating and uncomfortable challenge: productivity is clearly improving, but it is becoming much harder to measure accurately.
AI has changed how we write software. But more importantly, it has changed how we communicate intent.
Early conversations about AI-assisted development focused heavily on prompt engineering. Developers experimented with phrasing tricks, formatting styles, and clever instructions to coax better outputs from language models. Entire communities formed around "the perfect prompt."
That phase was useful, but it was never the destination.
Powered by Microsoft Foundry
👋 Hello Friend!
You can ask me about: