Mediations #35: Learning. Again. And Again.
Not only learning how to use AI, but also how a team works going forward.
Last week, I saw a message from December 18th in our team’s Slack channel. It mentioned a minor bug that affects only a few users. The team responded, “Let me create a ticket, and we’ll prioritize accordingly.” You can guess the story; the ticket stayed at the bottom of the backlog for almost two months. A bug affecting only a handful of users is often ignored and never prioritized in big companies with millions of users.
Luckily, the person posted a follow-up message last week asking whether there is any progress. Instead of responding with “we couldn’t get to it yet,” I triggered an Al agent as a response to the message.
I gave the AI (Cursor in this case), right from the exact Slack thread, ~30-word prompt saying, “@Cursor this thing X is connected to Y via A, and there is a fallback to B. But X occasionally doesn’t work as it should. Find the problem and fix it.” Within fifteen minutes, it found the issue and opened a pull request with a fix.
I don’t know why I was surprised, but I was. If I had approached it traditionally, I would have prioritized the bug ticket, refined it in the refinement meeting, then an engineer would pick it up, work on it, get a code review, and then merge the pull request. This process would take at least two days. It took AI to fix in fifteen minutes.
After the AI opened a pull request, we reviewed the code with one of the engineers on my team. The logic was spot on. We did one more iteration on the pull request (also with AI) to improve the code. Then, we merged the pull request, deployed and released the fix.
Keeping in mind that I work in a large company (3000+ people) operating in a regulated Fintech business, this one-thread conversation was not an easy destination to reach. But it was enough proof that we were moving in the right direction.
One prompt in a Slack message wouldn’t work if we hadn’t invested in the foundations up front. We are slowly integrating AI into our daily work. Recently, we refined AGENTS.md files and rules files, and advanced our AI code reviews to ensure Al agents have the right context. That Slack thread was not actually my first attempt.
My first attempt failed a few months back before we invested in defining contextual files for AI agents. Now that we have the context, I want to take the next steps with my team. Although I’m following people who are deep into agentic coding, I am approaching it with more mindful steps because we’re fundamentally changing how we work.
Meanwhile, and coincidentally, I was exchanging emails with one of my blog readers about how the engineering role has been changing for a while. We acknowledged that not only does the way of writing code change, but also how a team works.
How does a team review code now, or do they even review at all? How do they host daily standup meetings, or do they even need them? Do they even need sprint planning? All of them need to change.
We, as software engineering teams, must learn everything again. And again. We must invent new ways of working, form new team structures, and create processes that are different from what we know now. Right now, I don’t know what these processes will look like, especially in mid- to large-sized organizations.
The good thing is that we don’t start from scratch.
All technological advancements have built on top of existing ones. What looks revolutionary is almost always the accumulation of decades of small steps. Tiny pieces come together over time to enable a breakthrough.
We have at least two decades of agile practices, many more decades of software engineering, and even more decades of running business practices under our belt. Now it’s time to experiment with new ways of working. Take a step to make one process easier and better, then fail. Learn from it and then take another step. At some point, we will hopefully figure out a new way.
Perhaps we’ll need AI agents to communicate with each other instead of hosting daily standups, so agents don’t override or block each other’s work or create conflicts. But for now, we will keep suffering through the daily standup (as if we haven’t suffered enough) and take small steps to adjust existing meetings.
I must admit that I am nervous. Because of the future of my job, my profession is less predictable than five years ago. Although I’ve seen this coming (I have been closely following deep learning developments since 2017). I had high hopes of working in language models or machine learning, but I wasn’t brave enough to switch to a completely new profession back then due to life constraints.
That said, I’m not scared. Although I can’t predict the future (unlike many others who claim they can), I know I need to add many, many maaaany new tools to my tool belt. It’s time to learn and adjust. And then learn again.
It’s also time to ignore the voices such as, “Al will eat the world” or “Al can’t do my job, it slops...” I think both sides have a point, but also unhelpful.
The correct attitude is to accept the change, experiment with new tools and ways of working and learn from them. The right way is (and it always was) to solve the right problem at the right time. Only then can we focus on building products and systems that benefit people and help us make a living.
P.S. If you already tried new ways of working in a team with a mix of AI agents and people, hit reply to this email and let me know.
Good to Great
I share up to three things I found interesting, sorted from good to great.
Good: The debate between The Builder and The Thinker hit home for me. “I miss thinking hard,” says Ernesto. I feel you, Ernesto.
Better: There are people who affect your career trajectory. They change how you think, work and live. They are often unaware of what they have done, which makes it even more beautiful. I hope you will come across one (if you haven’t already). Susam did. Also, I admire Susam’s passion for computing spanning over twenty-five years.
Great: I used to think that I could control or influence a lot of things around me, especially in my job. When I couldn’t, I complained. In the last year, I changed my mindset around how I work. It was neither apparent nor sudden; it happened gradually. I am a much happier person now. Last week, I read this and realized that what Satya (the CEO of Microsoft) says is what I didn’t have before this change: an attitude of solving problems without complaining. Read it. Dissect Satya’s words, letter by letter.
Do you have suggestions for great things? Send me a message.