This is an article I wrote for LinkedIn in November 2025, published here for posterity.
I initially counted myself among AI skeptics because a lot of what was being said about this technology simply seemed too good to be true; my critical reflexes kicked in. Now, I use generative AI tools almost every day, and they’ve massively improved both my individual productivity and the quality of my work. As it turns out, I was simply using these tools the wrong way, preventing me from reaping the full benefit.
My mistake was trying to use AI tools to produce the bulk of the work for me, and then reviewing the results, making corrections where needed. Not only was this more mentally taxing, my productivity actually dropped because I had to spend so much more time reviewing and correcting. I had essentially used AI to just shift the work I had to do myself to be more difficult and time-consuming, without actually delivering better results at the end.
The solution is obvious in hindsight. When I started using AI to review what I had written and suggest improvements, my initial skepticism quickly faded. I started treating AI as an editor, not a ghostwriter. After all, the writing itself was never the hard part. Almost anyone can produce a great quantity of code or text, especially if they haven’t set a high quality bar. AI tools are great for mundane, repetitive work, but anyone who is doing that in software engineering – or writing – is probably doing it wrong.
Let me share some things that I find boring and repetitive: Combing through thousands of lines of product documentation to confirm one critical detail. Reviewing a suite of unit tests to try and find the missing edge cases. Re-reading and re-reviewing a document five times over to try and really make it shine. Now, tasks like these have been made much more expedient, and issues that might previously have been resolved by taking up a colleague’s time, I can now handle myself.
I found an issue with this approach. Sometimes I need to write code to do something that I don’t already know how to do, and it can be tempting to just have the AI generate it. After all, people have been sharing solutions on sites like StackOverflow for a long time, so what’s the harm? But again, this actually ended up a net negative for me because I’m again left with more of a hard problem to solve: that of accountability. My first duty is to the customer, and I hold myself accountable for meeting their expectations as well as I possibly can. To do that, I must know what I’m delivering to a considerable level of depth.
And so instead, I partner with a helpful colleague who has encyclopedic knowledge of the specific technology I’m trying to use, will instantly step to help me whether it’s 02 or 14, will never complain that I’m wasting his time (not that any of my human colleagues would!) and will tirelessly teach me what I need to solve the problem myself. This way, I can confidently hold myself accountable for delivering to the highest standards.