It all started with a conversation Sree initiated — probably in the office, in one of those rooms where you sit around a table and someone asks a question that sounds simple but isn't.
The question was: how can we start using AI? How can we start leveraging it?
Being a product person — someone who cares about agile ways of working, outcomes, incremental change — I heard it as a problem statement. AI is there. What do we do now?
And I had an answer. The wrong one.
I said: there are other levers we have in the organisation that will help us deliver better and faster. I said: it's too premature to think about using AI at this level to do anything interesting. I said: it might just be hype.
And I believed it, because I had lived through the hype before. The early internet boom. Big data. Mobile. Crypto. Each wave had come with the same breathless certainty — this changes everything — and each wave had been... complicated. Real, but complicated.
So while AI felt interesting, I was sceptical about how fast, how soon, it could help us in a structured way. We could have better discipline discovering problems, qualifying them, building software better and faster — without depending on something like AI.
That was my position. And I held it. Fairly loudly.
The conversation got heated.
And then I went home.
That evening, reflecting on what had happened, something shifted. I realised I had not listened to what Sree was actually asking. I had been so busy sharing my opinion — the other levers, the other problems — that I had completely missed the point.
All Sree was asking was: while we're busy with the day-to-day, can we start wondering what this new thing is all about? Can we start asking what it might mean for us?
Not: build a project. Not: commit to a goal. Not: deliver an outcome.
Just: wonder.
I messaged him that same night. I said — is this what you wanted? Not a solution, not a project, not a goal. Just a space for us to start asking what the art of the possible is?
He said yes.
And that set the journey for me.
Once that clarity landed, I moved from seeing AI as a next-word guessing machine — smart-sounding but fundamentally dumb, not really understanding anything — to asking a different question: if it can do this, what else can it do?
The answer came through a Chrome plugin.
As part of my role, I was heading the conversion rate optimisation programme. We used a tool to pull CrUX data — Chrome User Experience data from Google — but the tool had limited licences and I was always dependent on someone else to get me the numbers. The data itself was available. I just needed an API key and a bit of code.
I'd been putting it off. Modern programming means installing a hundred NPM packages, setting up environments, the whole scaffolding. It never felt worth it for what was essentially a simple data pull.
But I love Chrome plugins. Simple JavaScript, runs in the browser, no scaffolding. So I thought — why not build one?
I opened Windsurf. Gave it about six or seven lines of requirements. What I wanted. What the plugin should do. Nothing more.
In the next fifteen minutes, it generated working code. Five more minutes of debugging and it was live.
That itself was a phenomenal moment. Seeing a machine interpret what I wanted, produce readable code, and have it work.
But then the interesting part happened.
Once the basic plugin was working, I wanted to show benchmark data alongside the actual numbers — so you could see not just where you are, but what good looks like, and how far away you are from it.
So I prompted the AI to add that.
What it did surprised me.
It didn't just add the benchmark data. It reasoned about it. In real time, it argued — not just executed — that this was supporting information, not primary information. That it should be shown next to the actual data, but subtly. Not prominently. As context the user can see if they want it, but not so loud it becomes a distraction.
It adjusted the UI accordingly. Without being asked about design principles. It just reasoned its way to the right answer.
That was the magical moment for me. Seeing the context with which it interpreted what was being built, reasoned through what mattered, and modified the code to reflect that reasoning.
This was way before the models became as powerful as they are now.
And at that point, I was a convert.
Sree had latched onto this much earlier than I could see. And I had spent an afternoon in a room arguing with him about it.
This is not the end of the story.
The next chapter is building a full web app — which not only deepened the wonderment I found through the plugin, but helped me bring a few of my very smart, very sceptical technical friends along.
That one's coming.