At Microsoft Build this year, I had the chance to sit down with Cassidy Williams, GitHub’s Senior Director of Developer Advocacy, to dig into something that’s been on my mind: how agentic coding is fundamentally changing what it means to be a developer.
The conversation was eye-opening. We’re not just talking about AI writing code faster anymore. We’re talking about a seismic shift in where developers spend their mental energy, and honestly, it’s creating new challenges that nobody’s quite figured out yet.
The Great Shift: From Implementation to Strategy
For decades, a huge portion of developer work has been tactical: write the function, handle the edge cases, debug the weird interaction between two libraries. These tasks required deep focus and were often where junior developers learned their craft.
With agentic coding, that’s changing. AI agents can handle increasingly complex implementation tasks, which theoretically frees us up for higher-level work: architecture decisions, system design, business logic that actually matters. On paper, this sounds amazing. In practice? It’s messier.
What we’re seeing is that as the tactical burden lightens, the cognitive load actually shifts rather than decreases. You’re no longer thinking about how to implement something; you’re thinking about whether the agent understood your intent correctly. Did it make the right trade-offs? Is this the approach you would have taken? These meta-decisions are their own kind of exhausting, and they’re different from the satisfaction you get from solving a hard technical problem.
Decision Fatigue Is the Real Challenge
Cassidy touched on something crucial during our chat: decision fatigue. When you have an AI agent that can generate multiple approaches, write the documentation, suggest the architecture, you suddenly have more decisions to make, not fewer. Every suggestion is a branch point. Every option requires judgment.
This is where I think the industry is underestimating the human cost of AI acceleration. We talk about productivity gains, and those are real. But we don’t talk enough about the friction of constant micro-decisions. It’s cognitively different from writing code yourself, where you’re in flow state, problem-solving in a linear way.
The developers who’ll thrive in this environment aren’t necessarily the ones who can code fastest. They’re the ones with good taste, strong opinions about design, and the experience to know when an AI-generated solution is clever but wrong for the context.
Why Human Taste and Community Matter Now More Than Ever
This is where things get interesting for your career. If AI handles the implementation, what becomes scarce? Judgment. Taste. The ability to say no to a technically correct solution because it doesn’t fit your system or your team’s values.
Mentorship becomes more critical, not less. Someone needs to teach you how to evaluate AI-generated code, how to set good constraints for agents, how to think about systems at a higher level. This is traditionally been done through code review and working alongside experienced developers, and that dynamic needs to intensify.
Community feedback is also becoming more essential. When everyone has access to the same AI tools, what differentiates your work? Your opinions about what matters, shared through blog posts, open source, conference talks. I’ve written before about how building in public becomes more valuable as commodities become abundant.
The New Tools: GitHub Copilot’s Evolution
GitHub announced several updates to Copilot at Build, including a new GitHub Copilot app. What’s notable isn’t the features themselves, but what they represent: AI agents becoming more autonomous, more capable of working across your entire codebase, not just suggesting the next line.
This is the frontier. These tools are getting better at understanding context, maintaining consistency across a project, and handling more complex workflows. Which means the implications I’m talking about here will only accelerate.
What This Means for Your Next Move
If you’re early in your career, focus on developing taste and learning to evaluate quality, not just learning how to implement things. If you’re mid-career, think about how you position your expertise: what judgment calls can you make that an agent can’t? If you’re leading teams, consider how mentorship and taste become your team’s competitive advantage.
The future of development work isn’t about who can code the fastest or who knows the most APIs. It’s about who can make good decisions under uncertainty, who understands the trade-offs, and who can guide AI systems toward the right solution rather than just the working one. That requires experience, community, and something that still can’t be automated: good taste.
How do you plan to adapt your own work to this shift, and what aspects of development do you think will actually become harder to automate than the implementation itself?