For most of my career I've watched the same story play out every few years.
A new tool arrives. Sometimes it's a game engine, sometimes a project-management platform, sometimes a new way of doing agile. More recently it's AI assistants and "agents". Each time, there's a phase of excitement where the tool looks as if it will fix everything that was hard before: planning, communication, delivery, even strategy.
And each time, after the dust settles, the pattern looks familiar:
- teams with clear thinking, good judgement and solid habits get more effective
- teams without those things get a new way to make the same mistakes, faster
The temptation is always the same: to let the tool stand in for skills we never really built. AI just makes that temptation sharper, because it can now produce things that look like the output of those skills: strategy documents, code reviews, user research summaries, job applications.
The blunt version of my view is this:
Tools are leverage on judgement, not substitutes for it. AI is powerful leverage, but it doesn't change the basic equation.
If you don't have the skills to evaluate and use what it produces, you are not "working smarter with AI". You are outsourcing your thinking to something you cannot actually judge.
The old pattern in new clothes
Before AI, I saw this most clearly in game development and software engineering.
A new engine appears. It promises to make everything easier: better graphics, faster iteration, fewer technical headaches. It comes with demos and example projects that look impossibly polished.
Very quickly, you can spot three kinds of teams:
- Teams with strong fundamentals: good design instincts, clear production practices, healthy relationships between disciplines. They use the new engine to do what they already do, but with more reach.
- Teams that treat the engine as a silver bullet. They never really clarify what they're building, for whom, or why. They rely on the engine's feature list to define the product.
- Teams that get stuck in the tools themselves. They endlessly rebuild systems and workflows, and mistake internal neatness for value to players or users.
I've watched similar versions of this with version-control systems, methodologies, ticketing tools and presentation tools like PowerPoint or Keynote. The promise is always that the tool will fix problems of clarity, communication and decision-making.
What actually fixes those problems is people learning to model, prioritise, trade off and communicate, then choosing tools that support that.
AI hasn't changed that. It has just moved the pattern into domains people thought were "above tooling": writing, analysis, strategy, judgement.
Why AI feels so magical
One reason AI feels miraculous to many people is uncomfortable but important:
AI is doing work they never learned how to do themselves.
If you've never really learned to:
- structure a piece of writing
- critique a design
- read a contract
- sketch a strategy
- review code beyond "does it run?"
then a system that can produce something that looks like the output of those skills is astonishing. It feels like cheating in the best possible way.
The problem is that if you don't have those skills, you also don't have the ability to critically evaluate the output.
You can't easily tell:
- when the model has quietly changed the question
- when it has invented "facts" to smooth over gaps
- when the structure looks clever but doesn't actually answer the brief
- when the tone is subtly wrong for the audience
- when a code suggestion works but imports technical debt you'll be paying down for years
The risk isn't that AI makes a mistake. Everything and everyone does that. The risk is that it makes mistakes you can't see, and that you get used to trusting it anyway.
AI as amplifier, not brain
The way I've found most useful to think about AI is as an amplifier.
- Give it well-defined, bounded work, and it will do that work at inhuman speed.
- Give it clear constraints and good raw material, and it will surface patterns and options you might have missed.
- Pair it with strong human judgement, and you can explore more ideas, more quickly, than before.
But it still doesn't know what good looks like for you.
When I use AI in my job search, this is exactly how I treat it:
- It summarises long job descriptions.
- It compares a role with my CV and highlights matches and gaps.
- It creates a first pass at a structured answer that I then rewrite in my own voice.
At no point is it deciding what I apply for, how I position myself, or what trade-offs I'm willing to make. Those are questions about goals, values and context. They belong in my head, not in a model.
The same applies in team settings. AI can help:
- assemble data for a product review
- draft different ways of framing a strategy
- sketch test cases
- propose refactorings
It should not be deciding which customers matter, what risks are acceptable, or what kind of organisation you want to be. It doesn't live with the consequences of your decisions. You do.
The missing skills
This is the part of the conversation that tends to get skipped. If AI is going to be everywhere, what becomes more important, not less?
A few candidates:
1. Framing the problem
Being able to slow down and ask: what is the actual question here?
Most of the worst AI outputs I've seen are not "wrong answers" so much as "answers to the wrong question". That's not the model's fault. It was never told what mattered.
Framing is a human skill. It is a blend of domain understanding, empathy for the people affected and experience of what goes wrong.
2. Sense-checking structure
Even without specialised expertise, you can often tell whether something is well-structured or not:
- Does this strategy actually explain the problem, options and trade-offs?
- Does this application answer the questions that were asked?
- Does this "summary" surface what's surprising, or just repeat headings?
AI can propose structures. Humans still need to say "this doesn't hang together" or "you've skipped the hard part".
3. Having your own view of the domain
Every useful tool embeds a model: this is how projects work, this is how writing works, this is how code should be organised. If you don't have your own view of the domain, the tool's model quietly becomes yours.
With AI, this happens very fast. If you always accept the first suggested structure for a document, the model's habits become your habits. If you always accept the first code suggestion, its trade-offs become your trade-offs.
Having your own view doesn't mean being right about everything. It means caring enough to have an opinion, and updating it on purpose rather than by accident.
4. Developing a critical voice
AI is very good at producing text that sounds like everyone and no one. It averages.
Learning to have your own voice, in writing, in design, in code, becomes more valuable when average, generic output is cheap. It's the difference between:
- "This sounds plausible, ship it," and
- "This isn't how I would say it, and here's why."
That critical stance is not negativity. It is responsibility.
"It's not about the tool" as a design principle
When I say "it's not about the tool", I'm not being nostalgic or anti-technology. I've spent decades working with and benefiting from powerful tools. I like them. I build systems around them.
What I mean is that tools should be chosen and used in service of the way you want to think and work, not the other way round.
With AI, that design principle suggests some practical questions:
- What decisions in this process must remain human?
- What skills do we want to deepen, rather than outsource?
- Where can AI reliably save us time without distorting our judgement?
- How will we review and integrate its output back into systems we control?
In my own job search, I've answered those questions by:
- keeping a structure for my thinking that I own
- treating the search as a product system with a clear flow
- plugging AI into that system as an assistant, not as a brain
Other people will make different choices. The important thing is that those choices are made deliberately, not left up to whatever the latest tool makes easy.
The opportunity and the cost
AI genuinely changes what's possible. It's not just hype. There are things I can now do in minutes that once took hours.
But the opportunity is not "replace skilled people". It is "give skilled people more reach".
If we treat AI as a replacement for skills we never had, we'll get faster mediocrity and harder-to-spot errors. If we treat it as leverage on judgement and craft we actually care about, we'll get something worth having.
Either way, the tools will keep coming. Today's models will look quaint in a few years. The question that will still matter is the same one that mattered with engines, editors and project trackers:
What do you bring that is worth amplifying?
If the answer is "a clear way of thinking, built up over time, and a willingness to use tools without giving away my judgement", then it really doesn't matter what the next shiny thing is called.
It's not about the tool. It never was.