What Changed: MCP and Claude Code

Tue, 18 Aug 2026 08:53:04 +0100

A year ago I posted about an AI fail — Copilot confidently referring to code that didn’t exist while I was asking it to write Markdown. That post is the last thing I published here.

Since then something shifted, and it has become the norm. Almost forgetting what a pre AI day to day felt like.

Where I was

Mid 2025 was when AI started to feel real. But it just wasn’t quite good enough yet to be truly useful. The models that existed wrote basic code, but weren’t yet trusted by engineers enough to stop writing code manually or heavily scrutinising it. General chat capability felt useful to do research and play around, but again generally it didn’t seem trustworthy enough to use it in replacement to the traditional methods.

What actually changed

The models didn’t just get better at talking. They got access to the things I was talking about.

Towards the end of 2025 and into the beginning months of 2026 the models released became better and better. The output was better. It was starting to be trusted by engineering teams as code that could be used in production. Chat conversations seemed to be coherent and better quality. They began replacing the old fashioned Google search when wanting to do any kind of research. Pair this with mobile apps like Gemini and Copilot with access to your files and camera and you have what used to look like a sci-fi personal assistant.

Not just for coding but for general business use - I still think there’s a long way to go with this one, but the last few months have again got better and better functionality in this area. AI chat for your day to day applications like Microsoft Office, Teams, Outlook. Unfortunately maybe this one went a little far in every SaaS product wanting some kind of “AI” functionality whether it be useful or not.

AI agents started getting MCP (Model Context Protocol) support - “A standard way to hand a tool the things it needs to look at, so you stop being the copy-paste layer.”

This opened a door to the wider world for the context of the agent. It meant that you no longer have to manually copy and paste context or provide it with some other mechanism. The agent just has access to the systems and data that you’re wanting to work on/use. My biggest MCP use case is Azure DevOps for the creation, progress and closure of Work Items. It allows me now within my code editor to reference all my success criteria and tasks required for this specific work and have the agent go ahead and work on it as I would have manually.

Three things I do with it now

Write Code/Infrastructure as Code
Refine/create/execute Work Items within Azure DevOps
Use as a Google search replacement. Most research and investigations are now performed using AI chat

Where I’ve landed

I currently use Claude Code in my day to day work for most tasks. Outside of work I use Gemini and Claude quite heavily. We are at a point now where these tools are genuinely providing value and positive impact. We are seeing the shift towards AI writing more and more code. As with most of my IT career this is just another example of keeping up with the latest and greatest trends.

AI on The DevOps Quest

What Changed: MCP and Claude Code

Where I was

What actually changed

Three things I do with it now

Where I’ve landed