Thoughts on GitHub Copilot

I had originally written this post to express my disappointment at GitHub Copilot. That version came down to one paragraph:

GitHub is core to how I (and just about everyone else I know) build software. They had a chance to own this AI space, every repo, every workflow, every developer already in their ecosystem. Microsoft’s money, Microsoft’s reach, the infrastructure to ship to millions with a single update. They had the first mover advantage, but didn’t capitalise on it.

But since trying Copilot Coding Agent my opinion has changed.

This one feels like a leap rather than a minor update. The agent chains actions together, holds context, and stays with a whole task instead of helping with a block of code and disappearing. It carries a thread from start to finish.

The gap between tool and partner is shrinking fast, and the models will keep getting better. Some tasks that used to take an afternoon are done before you’ve finished your coffee.

Who knows what a team even looks like a year from now? Maybe it’s a mix of humans and agents, handing work back and forth. Maybe the “typical” team of 1 PM, 1 UX, and 5 engineers isn’t so typical for long. If the agent keeps getting smarter (which it will), the shape of the team shifts too. Maybe it’s fewer people spending more time on what matters, or maybe new roles appear that we can’t even describe yet.

This looks like the start of another shift in how teams work, where less time goes on glue and more goes on the product itself.


Published on .