December again. Snow outside the window in Calgary, tea on the desk, and the yearly habit of looking back. I do this review every year on the blog, and 2025 deserves a careful one, because it was probably the year my work changed the most since I moved to Canada.
The job change
The biggest personal event was leaving Benevity in April, after six years, to join Mawer Investment Management as a Senior Software Engineer. The short version is that I wanted a role closer to the product and the business, and that is exactly what I got.
Eight months in, I can say the bet paid off. My days are a mix of talking to stakeholders, turning what they need into requirements and architecture, and then building it. Less time in big meetings about process. More time understanding real problems and shipping answers to them. For someone like me, who always cared more about value than about code beauty, this is the right room to be in.
Was it scary to leave a place after six years? Yes. I had history there, trust, people I liked. Starting over at 40 something, in a new domain, was uncomfortable for a few months. Investment management has its own language and I had to learn it like a new English. But comfort is not a career plan. I am glad I moved.
Agents became normal
The second big change was not mine alone. This was the year agents stopped being a demo and became part of daily work. Not an experiment on Friday afternoon. The normal way we build.
In January I was still mostly typing my own code with some autocomplete help. By December, for a good part of my tasks, I describe what I want, an agent writes the first version, and my job is to review it, test it, and connect it to the real world. The shift happened so gradually that I only see the size of it now, writing this post.
What surprised me is which of my skills became more valuable and which became less. Typing speed, knowing API details by heart, remembering syntax. Worth less now. Writing a clear spec, smelling a wrong abstraction in review, knowing the domain well enough to spot a quiet logic error. Worth much more. I wrote about the spec part in October, and I believe it even more today. The bottleneck is thinking, not typing.
Review became a real skill this year, maybe the skill. When a machine produces code fast and with full confidence, the human who can read carefully and say “this part is wrong and here is why” is the one protecting the product. I spend serious time reviewing now, and I stopped feeling guilty about it. Review is not overhead. Review is the job.
What did not change
Here is the part of the year that I find most interesting. With all this change, the things that decided success and failure in my projects were the old things.
Understanding the user before building. Naming things by their business meaning. Keeping modules small with clear boundaries. Measuring before optimizing. Taking small steps and shipping often. Every one of these ideas is older than my career, and every one of them mattered in 2025 more than any tool choice.
The agents amplified everything. Good direction got amplified into fast progress. Bad direction got amplified into fast garbage. The fundamentals are what point the amplifier the right way. I keep going back to the writing of people like Martin Fowler, and almost nothing there expired. The advice about clear design and small steps reads like it was written for the agent era. It was just written for the thinking era, and that era did not end.
The money and risk view
I always try to close these reviews with the business lens, because this blog was never only about code.
This year, AI saved my team real time. Tasks that took days take hours. But the risk profile changed too. Fast wrong code is a new kind of risk, and the cost moved from writing to deciding and checking. Teams that understand this are budgeting for thinking time, spec time, review time. Teams that only celebrate the speed will meet the bill later. I work at an investment firm now, so maybe I am biased, but everything looks like risk management to me these days.
Hopes for 2026
I do not do big predictions. The last years taught me humility about that. But I have small hopes.
I hope to get better at the domain, because domain knowledge compounds like interest. I hope to write here more often, the months I skipped this year I missed it. And I hope we, as an industry, spend 2026 talking a little less about which tool is best and a little more about what we are building and for whom.
Thank you for reading this year. This blog is four years old now, started in a different world, and somehow the conversation only got more interesting. See you in January.
Pax et bonum.