What A(i) Year
What a year! What an AI year! What A(i) year. It's been a pretty unique 12 - 18 months for me - three jobs, a bunch of time off work and an AI revolution happening all around us. With all the change and being a regular new starter it's made me more open to experimenting with AI. There's another world where I'm still managing a codebase I'm v comfortable with and am a bit bemused by all the AI news. As it is I feel that I've got pretty lucky in the last year and have really enjoyed getting to grips with the technology. The models and tooling have also changed dramatically and become way more useful - so maybe in that other world I'd be using it just as regularly.
Either way this feels like a pretty unique time for me and the wider tech world so wanted to write a few notes of reflection.
No time, no money, poor results
At the end of 2024 my time was coming to an end at a small startup I'd been helping build for about three years. Lots of big ideas but little extra time or money. AI surely would help - however the models and tools back then were either constrained to simple autocompletion (and wow was that amazing when copilot first suggested full reduce function for the first time!) or were complicated tooling sets or expensive. We used what we could but a step change would involve both time and money without clear obvious results. While there was some hype it felt quite focused in a few places.
Time off / reset
In early 2025 I had a few months out of the tech world. We all know a reset is good and I think this chunk of time away from computers was good for me. It also coincided with big developments in the AI world. Agents were improving, claude code got introduced in Feb and rolled out officially at the same time I started a new role.
Time, money, good results
May 2025 I started a new job. When you've been at a place for a while you have a sense of everything, you're in control - especially when part of a small team and codebase. Starting somewhere new means those certainties aren't there. There was also encouragement to use AI, budget to do so and support for time to experiment. I got lucky starting the role as claude code got a wide release - it was the thing i was looking for - the terminal based interaction felt the most comfortable for my day to day work and the newish agentic workflow expanded things way beyond the day to day autocompletion.
Being somewhere new the AI was super helpful at researching the code base, helping me iterate and improve. I've since seen stats of how AI has got new starters up and running quicker and I can attest to that. It wasn't plain sailing - writing claude.md files felt odd - who knows if they work? The agents could be super persuasive and often very wrong. But like with most tools using it more and more you got a sense of where it would work or not. For example we wanted to generate some test data - the argument was that the AI would hallucinate ids etc. having played with it a bit the better route was to prompt it to create a script to generate the data - this could be clearly rationalised, improved and rerun. It's a pattern I've used regularly - but if I hadn't had some experience playing around I might have pushed it to generate some data and then been frustrated if it wasn't repeatable or properly structured. I definitely see a few people expecting miracles straight away without putting in some time to learn.
Headspace
Time and budget were super helpful (was nice to be at a place where money wasn't so tight - accidentally spending $500 on agents one month wasn't an issue). But I also think a big change was shifting from a hands on leadership role to be a full time individual contributor. I love working with others and helping them progress - but it does take up a lot of time and energy. Shifting to an IC role gave me a huge amount of free headspace to try things out, think, research etc.
As the agents have progressed and got smarter some of the skills I learnt as a manager came into effect. If you're used to managing a team you're used to things happening one step removed - you know what's going on but maybe not all the detail. You guide the team and nudge them. You're used to things sometimes going in the wrong direction, having to change tack, learning from mistakes. You have to be good at top level design and plans. Adept at having lots of tasks and projects running in parallel. Being a product focussed engineer also helped - how to keep the feedback loop tight and make platform decisions as one goes.
Another change - more structure / better models
A second new job in October 2025 - this one there was a huge remit and a v small team. There was also a v strong code base, lots of tests and great guardrails. At the new role there was also v clear structure of very small PRs and commits. Having this rigour on top of the AI skills I'd learnt was a huge boost. At the same time the models were getting better, we wrote better MD files, I got better at using the tools - and the tools and models got better. So much so that I rarely use my beloved code editor and live mostly in claude code.
The other path
I can imagine a path where i was working on the same codebase as a year ago - with tight time and money plus a role managing people. Am sure I'd be using AI still but I don't think I'd have had the time or headspace or the challenge of tackling a new codebase to progress as quickly. I can see how people are getting huge productivity out of AI having done it myself but can imagine a different world where I was less believing of some of the claims, where I had only dabbled but not had the time and support to learn properly.
How to learn
Like any skill it takes time. AI is doing the work you did before but maybe in a different way. The more you play with it you get a sense of how to guide it, where it works well, where it doesn't. You setup guardrails, plans, ideas, ways of working. But that all takes time, some money and the space to be able to experiment, to go slower before you go faster. Maybe that means pushing for that space at work, getting the support for training. Maybe it's sacrificing some of your own time or money. Or maybe you can wait. It's all changing so much - things that were useful or standard practice a year ago don't make any sense now. Hopefully you're in a supportive company that can provide time/money/training to help.
Conclusion
What a year! What an AI year! What A(i) year. Quite surprised at how it's transformed day to day work - I wouldn't quite have believed you just over a year ago that most of my day to day wouldn't be typing in a code editor but would be chatting to an agent in the terminal. About the only time I use my code editor is to write this blog! My year has been quite unique too - new jobs, time off and lots happening outside work. There's lots of fun at the moment, disbelief at some of the things one can build and iterate on so quickly and then either some existential dread about the future or bursts of joy and optimism about what one can make. Let's see what the next twelve months holds!