Quick prompts work fine when you need a small fix. But they break down on bigger projects. You end up rewriting the same instructions, adjusting them over and over, and hoping the AI stays focused. That cycle wastes time.
Microsoft just released Planning for Visual Studio Copilot in public preview. It’s now available in Visual Studio 2022 17.14, rolling out gradually. If you don’t see it yet, enable it through Tools > Options > Copilot > Enable Planning.
What Planning Does
Planning gives Copilot a structured way to handle multistep tasks. Instead of processing your entire request at once, it breaks the work into clear stages.
When you ask Copilot to tackle something complex, it evaluates whether to respond directly or create a plan. Simple questions get quick answers. Multistep tasks trigger the planning system.
Here’s what happens: Copilot creates a markdown file that outlines the task, defines research steps, and tracks progress. As it works, the plan updates. You can see what it’s doing, what’s finished, and what comes next.
The plan file lives in your temp directory (%TEMP%\VisualStudio\copilot-vs\). If you want to reuse it across different threads, move it to your repository. Microsoft is considering longer-term storage options in response to user requests.
How it Works in Practice
Planning uses transparent tool calls. You see the structure Copilot follows rather than guessing at its reasoning.
The system builds on hierarchical and closed-loop planning techniques. Copilot plans at a high level first, then executes step by step. As it learns more about your codebase or runs into issues, it adjusts the plan accordingly.
Progress tracking happens directly in the plan file. You always know where things stand.
One thing to note: If you edit the plan while Copilot is working, your changes might not take effect right away. Stop the response, update the file or prompt, then restart. Microsoft is exploring ways to make mid-response edits smoother.
The Results
Microsoft tested Planning against SWE-bench, a standard benchmark for evaluating AI coding tools. Both GPT-5 and Claude Sonnet 4 performed better with planning enabled.
The numbers: around 15% higher success rates and 20% more tasks resolved. These gains were consistently observed on larger, multistep problems where structure matters most.
Testing continues across additional models to confirm the pattern holds.
Why This Matters for DevOps Teams
DevOps work often involves complex, interconnected tasks. You’re not just writing a function. You’re updating deployment pipelines, refactoring services, or migrating infrastructure code.
Traditional AI assistants struggle with this scope. They lose context, make inconsistent decisions, or require constant hand-holding.
Planning changes that dynamic. When Copilot can research your codebase, outline a strategy, and adapt as it works, it becomes more reliable. You spend less time babysitting the AI and more time reviewing its output.
The transparency helps too. You can review the plan before Copilot executes it. If something looks off, adjust the approach early, rather than discovering problems after the code is written.
What Developers Need to Know
This is a preview feature. Microsoft is still refining how plans are formatted and how they evolve as newer models become available.
The current limitations are manageable. Plan storage is temporary unless you move files manually. Mid-response edits require stopping and restarting. These aren’t dealbreakers, but they mean you’ll need to adjust your workflow slightly.
Microsoft is actively collecting feedback. If you want to influence how this develops, share your experience through the official feedback channel: Planning/To-dos in Copilot Chat.
An Analyst’s Perspective
Mitch Ashley, vice president and practice lead, software lifecycle engineering, The Futurum Group, believes, “It’s important not to view Visual Studio Copilot’s Planning solely as a new standalone feature. Microsoft’s Planning feature in Visual Studio Copilot reflects the next step forward in the Agentic Development Lifecycle, which I refer to as Intent-Driven Development. Instead of reacting to prompt after prompt, AI is led by interpreting developer intent, creating structured plans, and executing multistep workflows with transparency and reasoning. It’s a shift from code generation to outcome creation, where developers guide the ‘why’ and AI manages the ‘how.’”
“Intent-Driven Development represents an important step forward in how software is created using AI. By setting the intended outcome in motion, planning becomes the bridge between the work that must be performed and agentic automation. Developers move from crafting code with prompts to shaping the larger solution they seek, supported by agents that are directed by goals and plans, adapt to context, and coordinate more of the software development lifecycle.”
What’s Coming
Microsoft plans to improve caching, refine Copilot’s reasoning, and deepen its understanding of project context.
These aren’t vague promises. The company is treating Planning as the foundation for planning-driven development in Visual Studio. The current release establishes the framework. Future updates will make it smarter and more integrated.
Should You Use It?
If you work on complex codebases or manage multistep refactoring, Planning is worth testing. Enable it, try a few tasks that normally require multiple prompt iterations, and see if it reduces friction.
The 15-20% improvement in benchmark performance suggests real value, but benchmarks don’t always match real-world conditions. Your mileage will vary based on your specific codebase and workflow.
The transparency alone makes it useful. Even if Planning doesn’t always produce perfect results, seeing how Copilot approaches a problem helps you guide it more effectively.
Start small. Test it on non-critical work first. If it saves time and reduces frustration, expand its use gradually.
Planning represents a different approach to AI-assisted development. Instead of hoping the AI understands your intent, you get a visible, adjustable roadmap. That shift toward transparency and control matters more than any specific performance metric.
