There’s a saying in triathlon: why be good at one sport when you can be bad at three sports?
Ask me how I know.
That line has started to feel increasingly relevant to how I use AI. I spent most of my career in analytics, which meant I was reasonably good at working through data, finding the real problem and explaining what it meant for the business. The moment the work moved into software development, interface design, deployment, infrastructure or writing production-ready code, I was outside my lane. I could understand what needed to happen, but actually making it happen usually meant finding someone with a completely different set of skills.
AI has changed that. It hasn’t made me an expert developer, designer, copywriter or systems engineer. It has made me just competent enough in all of them to keep moving.
That distinction matters. A lot of the conversation around AI is still framed as replacement: can it replace a developer, an analyst, a designer or a writer? I’m more interested in what happens when it lets one person cross the gaps between those roles. I can start with a business problem, work through the data, sketch the workflow, build the backend, put an interface around it, deploy it and explain what it does. None of those individual pieces would impress the best specialist in that field. The interesting part is that the whole thing exists.
Revisi is a good example. The core idea did not come from asking an AI to invent a startup. It came from a fairly normal problem: website copy grows across product pages, landing pages and blog posts, and eventually it becomes difficult to tell whether it still sounds like the same brand. I understood the problem because I had run into it myself. I could define what a useful review needed to look like, decide that vague “does this sound right?” prompts weren’t good enough, and structure the output around explicit voice measures and prioritised recommendations.
The parts I didn’t know were everything required to turn that thinking into a usable product. How should the crawler work? What should be cached? How should the scoring process be structured? What happens when only three pages have changed? How do I stop the interface from looking like it was assembled during a regional hackathon in 2012?
AI helped me work through those questions as they appeared. I still had to decide what made sense, test the output and notice when the proposed solution was ridiculous, but I no longer needed to spend six months becoming competent in every adjacent discipline before I could build anything. I could learn enough at the point I needed it.
That is probably the most useful version of AI for me. It reduces the cost of curiosity.
Previously, wandering outside your speciality came with a large time penalty. You could either stay in your lane or disappear into tutorials, documentation and Stack Overflow threads until you vaguely understood what was happening. Now I can explore an unfamiliar area, ask increasingly specific questions and get to a working level much faster. Not mastery. Enough understanding to make a decision, inspect the work and move to the next problem.
There is an obvious downside. AI also makes it easier to be confidently bad at things. You can create broken software, misleading analysis and dreadful writing much faster than before. A working button does not mean the workflow makes sense. Code that runs is not necessarily code you should trust. A polished answer can still be based on the wrong assumption.
Being bad at more things is only useful when you are good enough at one thing to recognise quality.
For me, that anchor is analytical judgement. I’m used to asking whether the question is properly defined, whether the data supports the conclusion and whether the result is useful to anyone. Those habits carry into product building. AI can suggest an architecture, but it cannot decide whether the product deserves to exist. It can produce a scoring model, but it does not know whether the scores reflect how people actually judge brand voice. It can build a dashboard, but it does not know which decision the dashboard is meant to improve unless someone defines that first.
The people who get the most from these tools may not be the ones who try to become experts in everything. They may be people with depth in one area who are curious enough to become temporarily competent in several others. An analyst who can build. A designer who can prototype the logic. A subject-matter expert who can test an idea without waiting for a full product team.
AI has not made specialists irrelevant. I still wouldn’t hire myself to redesign a banking platform, secure medical infrastructure or write the software keeping a plane in the air. Some work should absolutely be done by people who are very good at one sport.
But for smaller products, internal tools and ideas that would otherwise die between “someone should build this” and “we don’t have the resources,” being bad at three sports is suddenly pretty powerful.
It still hasn’t helped my swimming.