Working Full-time with AI

I wrote a post back in June that I titled “My New AI Co-Worker” and I talked about how this was completely changing my approach to software development. Well, I have good news, all of that is still true! Since the start of 2025 myself and my employer have fully leaned into using AI as a primary tool to start any initial execution of work.

I’ve been empowered to do more than I ever could have alone. I work faster, I work more effectively, and I can step outside my lane more often. All of this results in what I talked about in my last post, AI just becoming a massive force multiplier for any organization.

Being able to work through multiple problems, bug-fixes and other issues at one time is amazing. While I’m putting my effort in a specific place, AI agents can be working on other parts of the application for me. As I’m writing this I have two AI agents working on different problems for me so I can keep moving my work forward.

I’ve become a manager of robots.

Becoming a robot manager isn’t all bad… I mean, they rarely talk back, and they’re always ready to work when you are. But, in reality having these robots work through problems, then I can just check back in on them and see what they’ve accomplished when I’m ready is amazing.

But whats the catch? I’ve had to level up my own skills in unexpected ways. I’m writing more detailed specifications, because vague instructions lead to wasted time and prompts. This has the spillover affect of allowing me to break down all the work I’m doing into much smaller, more digestible chunks. I’m also doing more code review in a day than I used to in a week, and my sensibility for catching issues during this process has grown sharper. I’ve become much better at catching Architectural issues earlier because overall I’m thinking about my work at a higher level.

I think working with AI has started to make me a better engineer.

I think this has improved my working relationship with my human colleagues significantly. I’m able to take more time to have more thoughtful communication. I’m working through tasks faster and more effectively. I’m able to cross-discipline more often and take some additional burden off of teammates that would normally pickup and finish a particular piece of work.

Some problems are still better solved the “old way”

Not everything should be handed off to an AI agent, and learning where to draw that line has been crucial. Deep architectural decisions that require understanding years of technical debt, business context, and team dynamics? That’s still very much a human job. AI can help explore options, but the final call needs someone who understands the full weight of that decision.

Quick hotfixes often fall into this category too. When production is down and you need a five-line fix, the overhead of setting up an AI agent, writing clear specifications, and reviewing the output often takes longer than just fixing it yourself. Sometimes the fastest path is still your text editor and your expertise.

I’ve also found that debugging truly mysterious issues – the kind where something works in staging but fails in production, or where the bug only appears under specific race conditions – requires that human intuition and creative problem-solving that AI still struggles with. AI is great at checking the obvious culprits, but those head-scratchers that require you to form and test hypotheses? That’s where your years of experience shine.

And let’s be honest: delicate refactoring of critical systems is still something I want human eyes and hands on from start to finish. The risk is too high, and the need for judgment calls too frequent.

Advice for other engineers making this transition

Start small and specific. Don’t try to hand off your entire sprint to AI on day one. Pick a well-defined task – maybe a bug fix or a small feature – and learn how to work with AI on that. You’ll quickly develop a sense for what works and what doesn’t.

Invest time in learning to write better specifications. This is the skill that will make or break your productivity with AI. Think of it like writing requirements for a junior developer who’s brilliant but has zero context about your project. The clearer you are, the better results you’ll get. This skill has spillover benefits for working with your human colleagues too.

Never merge without understanding. I review every single line of AI-generated code. If I don’t understand why something was done a certain way, I either ask the AI to explain it, or I rewrite it myself. Treating AI output as a starting point rather than a finished product has saved me from several potential issues.

Maintain your core skills. Just because AI can generate code doesn’t mean you should stop learning how things actually work. I still do code katas, read documentation, and work through problems manually sometimes. Think of AI as a power tool – incredibly useful, but you still need to know the fundamentals of carpentry.

Finally, be patient with yourself and your tools. This is a new way of working, and you’ll have days where the AI seems to be fighting you at every turn. That’s normal. Over time, you’ll develop workflows that feel natural, and you’ll know instinctively when to reach for AI and when to just write the code yourself.

The goal isn’t to replace your skills – it’s to amplify them. And that’s exactly what’s been happening for me. Coding with AI has brought a renewed joy and interest to my work. Things have suddenly started to feel exciting again, and its only just begun.