Github Copilot: Has your new AI teammate arrived?
Don't stand still. I remember hearing this 15 years ago and thinking it was great advice when working in a technology field. It just made sense. With the fast pace of change and not standing still by always learning something, it avoided your knowledge becoming irrelevant. Fast forward 15 years to 2025 and I think I need to update this advice. It fails to convey the speed required just to remain relevant in the age of AI. I think the correct advice is that we should be 'jogging', somewhere between walking and running. In other words, not slow enough that you’re waiting for others to form opinions on the latest technologies on your behalf, and not quickly enough that your adoption of tooling is irresponsible or unsustainable. We need to be going at a brisk jogging pace just to remain relevant.
I say this as 2025 is shaping up to be a pivotal, transformative year in how software development happens, and we should already be jogging. The rise of AI-powered software agents, particularly GitHub Copilot’s newly released coding agent announced at MS Build this week, marks its shift from passive assistant to active participant in software development. It will change how we think about collaboration, productivity, and the very nature of coding itself.
The Evolution of Copilot: From Assistant to Agent
When GitHub Copilot was first introduced in 2021, it was billed as an “AI pair programmer”. A helpful sidekick that could autocomplete code, suggest snippets, and reduce the cognitive load of repetitive tasks. With the newly released GitHub CoPilot Coding Agent we're now entering an era of AI Agents capable of autonomously tackling tasks of a low-to-medium complexity. Be that bug fixing, writing features, refactoring code, or improving documentation, these agents can get to the outcome without a human actively prompting them every step of the way. Human in the loop has been the prevalent pattern until now with humans initiating and approving each action, but I see this now shifting to human on the loop with humans reviewing and supervising the output of coding agents.
Developers can today assign issues directly to the GitHub Coding Agent which spins up a secure development environment, analyses the codebase, and begins work in the background aligned with custom instructions matching your software development standards e.g. testing/linting. Once done, it pushes commits to a draft pull request, logs the actions taken with its rationale, and awaits human review. Following human review, the developer then has the choice to merge the work directly or intervene to correct something that’s went wrong.
A profound thought: Agentic DevOps
A new phrase again shared this week at MS Build was the concept of Agentic DevOps, describing this shift in the pattern as we move from a model of pair programming driven by human interaction with AI tools augmenting, to one of peer programming where the AI tools are given a problem and off they go to solve it. In many respects this feature, as impressive as it is technically, raises for me more significant philosophical questions. Chief among them is whether this is the point we ought to start considering agents as capable teammates, albeit, junior ones? They accept assignments, react with emoji acknowledgements and whilst there are many things that can and will be improved over time including the ability to learn from past interactions as a human teammate might, it’s now much easier to imagine a future where teams include both human and AI agents, each bringing their own strengths to the table.
Trust, Transparency, and Human Oversight
Of course, with great power comes great responsibility. With introducing this new Coding Agent GitHub have taken steps to ensure new capabilities don’t compromise security or quality. All changes are applied on a branch with any pull requests generated by the agent still requiring human approval before triggering CI/CD workflows. Branch protections remain intact, and session logs provide full transparency into the agent’s decision-making process.
I believe we will see a proliferation of this model where AI Agents will be included within teams to handle what developers might today consider toil, with humans then supervising its completion. In many respects this is not dissimilar to the original DevOps movement which sought to - through automation and other practices - reduce toil and free up human brain cycles for more innovative, rewarding and challenging work. So, with AI Agents deployed in this way to handle the toil of bug fixing or avoiding the need for developers to wade through package dependency conflicts, what do we then spend that time doing that fosters innovative, challenging, rewarding and impactful outcomes? This to me invariably leads me back to business value and developing a healthy obsession that links the two.
What This Means for the Future
The transition from pair to peer programming isn’t just a technical milestone; it’s a cultural one. It challenges our assumptions about what it means to create software, how teams are structured, and how work is distributed. It opens the door to new forms of collaboration, where humans and agents co-create in ways that were previously unthought.
For organisations, this means rethinking workflows, training, and even hiring. The skills we need in the future to create software will extend far beyond coding to prompt engineering, system design, and the ability to collaborate effectively with AI agents.
We’re living through the dawn of a new era in software development. Jog on.