GenAI Turned Two. Most Companies Are Still Writing Their First Prompt.

I remember exactly where I was the first time I used ChatGPT. I am not sure that is something people admit in polite company, but I will say it. It was late 2022, I had a strategy brief due, and someone on my team sent a link with just try this. I typed in a fairly dry competitive analysis question, mostly out of curiosity. The response came back in about eight seconds: structured, coherent, competent. Not brilliant. But competent at a speed no human team could match.

I sat back and thought: okay. This is different.

Two years later, I have a very different thought: most organisations still have not figured out how to use this well. I mean that less as a criticism and more as an honest observation of what the last two years have actually looked like from where I have been sitting.

2024 was the year AI went from a thing everyone was curious about to a thing every company felt obligated to have an opinion on. The press releases multiplied. The AI strategy decks circulated. The innovation committees formed. The working groups formed to support the innovation committees. The POCs started. Some of them are still in POC status today, which is a sentence that should make everyone at least a little uncomfortable.

The models got astonishing this year. GPT-4o brought voice and vision together in a way that felt genuinely magical the first time I used it. Claude 3.5 Sonnet became my go-to for almost everything involving writing and reasoning. Gemini started closing the gap in ways that were easy to underestimate. The capability race was real, the investments were real, and the winners were all of us, because the tools got dramatically better and dramatically cheaper.

What did not keep pace was integration. Not the technical kind: there are enough APIs and low-code tools now that connecting AI to existing workflows is genuinely achievable for most teams. The integration I mean is the harder kind: into how decisions get made, into what questions get asked, into the actual texture of the work. That gap, between what the technology can do and what organisations have managed to do with it, is the real story of 2024.

I watched organisations run impressive demos that never left the demo. I watched AI-first get declared in the all-hands and quietly de-prioritised by Q3. I had conversations with people who are genuinely smart about AI strategy and are still, eighteen months in, struggling to connect the technology to an outcome that changes how their business actually operates.

"The teams I have seen make real progress this year all started with the same question: what is the most painful thing our people do every day, and can AI make it meaningfully better? Not: how do we use AI?"

What I have also seen, increasingly, are the exceptions. Teams who started with a specific, painful problem and built something small that actually worked. And then the next thing. And found that the speed compounds in a way that broad exploration never does. These teams are not in the headlines. But they are the ones the headlines will eventually be about.

GenAI is two years old. The platform shift has happened. The capability is real. What 2025 needs to be is the year we stop asking should we use AI and start asking are we using it well enough. Those are very different questions. Most organisations are still on the wrong one.