Boosting DevOps Efficiency with GitHub Copilot

Recently set up Azure DevOps MCP in VS Code and integrated it with GitHub Copilot — and this changed my workflow more than I expected. Now Copilot doesn’t just suggest code in isolation… it understands the actual work item from ADO. Here’s what that looks like in practice: • Pick a user story / bug from ADO • Copilot reads the context (description, acceptance criteria) • Starts suggesting relevant code changes • Even helps with test cases and edge scenarios • I review, tweak, and accept The shift is subtle but powerful: Earlier: Understand requirement → design → write code → validate Now: Understand requirement → review + refine generated code The role changes from “writing everything” to thinking, validating, and guiding. Of course, it’s not perfect — but for well-defined tasks, it significantly reduces effort and speeds up delivery. Feels less like autocomplete… more like working with a context-aware pair programmer. Curious — would you trust AI to generate code directly from your work items? #GitHubCopilot #AzureDevOps #VSCode #AIinDevelopment #DeveloperProductivity

Agentic Platform Engineering is here We’re moving from AI-assisted coding → AI-driven execution. With GitHub Copilot + Azure: • AI agents can plan, build, test, deploy • Developers shift to orchestrating systems, not just writing code This is the next evolution of Platform Engineering & DevOps. Blog: https://lnkd.in/dqiuyYCt GitHub: https://lnkd.in/diu3jdad

The Terraform challenge: How do I build the most resilient and reusable Terraform possible with the least amount of third-party cost? Not just reusable code. Not just modules. A framework that lets teams consume infrastructure with minimal effort, full flexibility, strong versioning, and as little paid platform dependency as possible. Most teams go through the same progression: Phase 1: Flat Terraform (This is where most AI Generated terraform is) Phase 2: Modules with variables Phase 3: Object-base Modules Then, if they keep going Phase 4: Declarative execution Future: Platform and other capabilities The problem is most teams stop at phase 1 or 2. They end up with reusable code for the author, but not easy consumption for the user. AI is great at using what's out there which is rarely even modular (Phase 2). It has only been trained on documentation and the "hello world" of Terraform. The real goal, in my opinion, is bigger. Make Terraform: ✓ Reusable across projects and environments ✓ Easy for consumers to use without terraform experience ✓ Minimal or zero Terraform code for the consumer ✓ Work across vendors ✓ Versioned, consistent, predictable ✓ Flexible for security and business requirements ✓ Low on dependency extra tooling or scripts ✓ Consistent and reusable if using API's, AI, MCP It has to be able to have dynamically generated modules based on provider changes. Then reduce the consumer experience to structured data through custom solutions, not hand-written Terraform. That is when Terraform stops being a programming surface and starts becoming an execution engine. This brings Terraform back to the Declarative solution that was promised. This is the direction I think is worth building toward. This is the framework solution I have been developing and using. I'm going to be putting together some videos and calls to step through these phases of Terraform, its adoption, and this framework. #Terraform #IaC #DevOps #PlatformEngineering #CloudArchitecture #IACFoundry

After using Agent Skills with GitHub Copilot for the past week, the productivity uplift is very real especially for developers and architects. Unlike MCPs, which focus on retrieving information, Agent Skills focus on - ✅Best practices ✅Architectural patterns ✅Constraints and trade‑offs On top of that, you can create custom agents and add your own prompts, making Copilot even more grounded, contextualized, and aligned to your architecture, standards, and domain. This means - ✅Higher‑quality first responses ✅ Less re‑prompting and correction ✅ Copilot that reasons more like a domain expert, not a search engine MCPs and Agent Skills work well together but Agent Skills are what turn Copilot into a context aware/ domain aware partner. Read more here - https://lnkd.in/emAS_NA4 https://lnkd.in/eX7ymntK #AgentSkills #GitHubCopilot #DeveloperProductivity #SoftwareArchitecture #AIEngineering

What I Learned Building a Custom GitHub Copilot Agent to Triage Nightly Test Failures If your team runs nightly builds, you probably know the routine: arrive in the morning, see a wall of red test failures, and spend the next chunk of time figuring out which ones matter. Is it a real regression? A flaky test? An environment hiccup? I wanted to see if GitHub Copilot's Agent Mode could help with that — so I started experimenting. The result is a custom Copilot agent that: 🔍 Fetches test results from Azure DevOps nightly builds via the ADO MCP server 📊 Attempts to classify each failure (Regression, Assertion, Timeout, Flaky, Environment…) 📈 Generates an HTML summary report with pass/fail breakdowns 📋 Creates Tasks in Azure DevOps with error context and investigation pointers 🛠️ Suggests possible fixes by reading the test source, system-under-test, and stack traces 🧠 Tracks past fixes in a local vector store — so it can surface similar solutions instead of starting from scratch ⚡ Flags potentially flaky tests by looking at pass/fail patterns across runs It's far from perfect — the classifications aren't always right, and the fix suggestions still need human judgement. But it's been a useful learning experience in what's possible with Copilot's customization layer. The building blocks that made it work: Custom Instructions — give Copilot persistent context about your project, tools, and conventions Agent Skills — package repeatable workflows into reusable modules Prompt Files — wire up one-click entry points that chain skills together MCP Servers — let Copilot talk to external systems like Azure DevOps No separate backend needed — it all lives as configuration-as-code in the repo. What started as a learning exercise has turned into a working prototype — and early results suggest it could compress morning triage from the better part of an hour into minutes. Still iterating, but I'm excited about where it's heading. Would love to hear if anyone else is exploring custom Copilot agents for their own workflows. #GitHubCopilot #AgentMode #AzureDevOps #TestAutomation #AIEngineering #DevEx #DeveloperProductivity

This repo introduces simple, powerful, and architecture-first strategy for structuring Copilot agent projects based on best AI engineering practices - including plan first, delegate to sub-agents, pursue elegant form, fix what’s broken, validate everything, learn from mistakes - a truly fantastic template for reproducible agentic configuration 🤖⚙️! https://lnkd.in/dGHszBdt #githubCopilot #agentArchitecture #agenticAI #projectStructure

[New Blog Post] Drawio MCP for Diagram Generation: Why It’s Worth Using I’ve been using GitHub Copilot & Drawio MCP to generate diagrams directly from structured input - tied to code, infrastructure, and workflows. Instead of manually arranging shapes, I can produce a draft in minutes, refine it deliberately, and commit it to Git. 🔗 Read the full post here: https://lnkd.in/ec4Wchai #PlatformEngineering #Azure #DevOps #InfrastructureAsCode #DiagramAutomation #GitOps #AIinEngineering #GitHub #GitHubCopilot

The Draw.io MCP with a proper SKILL.md and a clear user tutorial with examples, it’s simply stunning 🤩 - a fantastic idea how well designed AI tool can improve and automate your work #agenticAI #platformEngineering #architecture

Most organisations are letting GitHub Copilot agents loose with prompts and hoping for the best. That’s not an operating model. That’s a risk! To go deeper on the last post, I shared a repo — a practical way to train, govern, and standardise how GitHub Copilot agents behave inside your codebases. 👉 https://lnkd.in/eRQh8ms5 The premise is simple: Instead of prompting harder, codify engineering discipline so agents execute with — consistently and at scale. 🧠 What the Dojo enforces (the “Six Disciplines”): 🥋 Plan before striking → Agents must plan the work before touching code 🧩 Delegate with sub‑agents → Research & parallel analysis without flooding context 🔁 Learn from every fall → Lessons captured after corrections so mistakes don’t repeat ✅ Prove the technique → Tests, logs, diffs required before anything is “done” 🎯 Pursue elegant form → Agents challenge hacks and shortcut fixes 🐞 Fix what’s broken, solo → Reproduce → diagnose → fix → verify, no hand‑holding 📜 What’s actually in the repo: • skills.md → the core behavioral “kata” agents auto‑discover • .github/copilot-instructions.md → the house rules and workflow and lessons.md → self-improvement This is the same shift we made years ago with: • CI/CD pipelines • Cloud landing zones AI agents are next. 👀 Question for tech leaders: Are you treating AI agents as first‑class team members with standards and training? Or are they still freelancing in your organization’s repos? Because unmanaged agents don’t just move fast — they move fast in the wrong direction. #AgenticAI #GitHubCopilot #microsoft #EngineeringLeadership #AIArchitect #DevEx

Are you measuring GitHub Copilot usage… or real impact? 🤔 Many teams proudly track seats, active users, and suggestion acceptance in GitHub Copilot. That’s adoption. But leadership conversations are shifting toward something else: 👉 Is AI actually improving engineering performance? This is where gh-devlake changes the game. Built as a GitHub CLI extension, gh-devlake connects Copilot usage data with real delivery metrics across your engineering systems. More details and how to start: https://msft.it/6048Q5AMC Instead of looking at AI stats in isolation, you can correlate: - Copilot usage - PR cycle time - Deployment frequency - Change failure rate - Mean time to recovery Now we’re talking about impact. What I like about this approach is that it moves the conversation from opinion to evidence. You’re no longer asking whether AI “feels” productive. You’re analyzing how AI adoption aligns with actual delivery outcomes in your organization. For engineering leaders, this is critical. AI investments need defensible ROI narratives. gh-devlake gives you the foundation to build them using your own DevOps data. We are entering the phase where AI in engineering is not about experimentation anymore. It is about measurable value. Are you already correlating Copilot adoption with delivery metrics? Or still tracking usage alone? Eldrick Wega, Morten Stange Bye, Haakon Hasli, Else Tefre, Christian Tryti, Majid Hajian, Pankaj Agrawal, Muhammad Daniyal (Dani), Chris Romp, Daniel Meppiel, Jonas Helin, Joel Norman, Adil I., Ömür Sert, Francesco Manni, Jaime De Mora #GitHubCopilot #EngineeringLeadership #DeveloperProductivity #DevLake #DevLakeCopilot #msftadvocate