Why AI can’t replace what you’ve spent decades building

The factory has finally reached the desk. When the power loom arrived, it didn't wipe out textile work. It wiped out the routine part, the weaving anyone with a week's training could do, and pushed the trade toward what the machine couldn't touch: judging the cloth, the pattern, the market.

Ethan Mollick warns the same industrialisation is now happening to knowledge work. Craft fields, he says, get pressured to produce a standardised product.

For you, that pressure is already here. It shows up as the client who asks why they should pay your rate when an AI can produce a "competent" version of the same thing for a fraction of the cost.

But "competent" is carrying the client's whole argument, without having earned it. What the machine produces cheaply is procedure, the repeatable steps. What it can't produce is knowing which of those steps is right for this client, in this situation, at this moment.

AI has made procedure cheap. It has not made understanding cheap.

That is the democratisation of expertise, and it is real. Richard and Daniel Susskind predicted it in The Future of the Professions: technology would let less-qualified people do work that once needed specialist training. AI has made their prediction come true faster than they expected. Legal research, medical triage, financial analysis, marketing strategy. Anyone willing to learn the interface can now produce a competent version of all of them.

But the Susskinds' thesis has a hole in it, and I think it matters for anyone who has spent a career building a practice.

What AI actually democratises

The Susskinds are right about procedure. A junior professional with AI assistance can draft a legal document, generate a marketing strategy, or analyse financial data to a standard that used to take years of apprenticeship to reach.

Filip Drimalka, a digital innovation specialist and the author of The Future of No Work, calls this symbiosis with AI. Not asking whether to use it, but treating it as an extension of your own thinking in every step of the work.

There are two ways to run that symbiosis, and it's worth being precise about which you are doing.

As a Centaur, you split the work and hand a whole task over to the AI, then come back and steer the draft. 

As a Cyborg, you and the AI work through it together, step by step, your thinking and its output intertwined.

I do both. Every month I write the SEO reports for my clients. The AI sets up a list of client folders and downloads the analytics reports into each one. One particular process cannot be automated, so I do that by hand – but right now I am working with my AI to figure out an easier way to do this than by hand.

The AI then processes and organises the reports, and produces a draft report that picks up anything I may have skipped. I then review the draft report. That is Centaur work.

The domain migration was Cyborg work. It talked me through each step and I checked each one before the next.

For procedural tasks, both work. The AI does the processing, you do the orchestrating, and the two together produce competent output for a fraction of the old cost. That is genuine democratisation. And for any profession that built its fees on procedure alone, it is genuinely disruptive.

What AI cannot democratise

Here is where the Susskinds' thesis breaks down. They assumed expertise is mostly procedure – rules, standards and processes that can be written down and handed over. For some professional work that's true. For the expertise that comes from decades of doing the work, it isn't.

That's the part of your expertise that matters most and works before you can explain it. It's tacit knowledge, buried deep down. You walk into a situation and your brain has already compared it against the many professional encounters you have lived through, and you have a sense of what will work before you can say why.

That sense can't be written down and handed to a junior colleague or a machine. It can only be built by doing the work, over and over, for years.

Tomáš Baránek puts it this way – the AI doesn't know anything, he says. It recombines what it has already been given, on probability principles. And the less you understand a topic yourself, the more likely you are to be fooled by a wrong answer and not notice.

That is the distinction in one line. AI can reproduce the words of understanding. It can't do the understanding.

I see the gap every day. I feed the notes I have taken into a model, and it hands me back things that don't make sense, things that don't relate to what I asked. So I have to push back until they do. I keep some of what it gives me, throw most of it away, and decide what goes out under my name.

Let me make this concrete, because it is easy to talk about working with AI in a way that sounds grander than it is. Every morning a note is waiting that I didn't write: the weather, what's on today, the reading I saved overnight, the posts queued for the week. I keep a shelf of skills for the jobs I do often, so I can ask for a newsletter or a check against my own voice and the steps are already there.

I now have a small library of several books loaded into Gemini Notebook. Each one becomes a coach when I paste a draft into its notebook and ask it to read as that author would – Steven Pinker's The Sense of Style for how I write, Ethan Mollick's Co-Intelligence for how I work with AI.

I also use a separate model that reads with fresh eyes and tells me what doesn't fit. When a job is big I split it into pieces that run alongside each other. And most days I feed it what I've been reading and ask for the shape of it: a summary, the themes, the questions it raises. The machine does the gathering and the drafting. I do the deciding, and to be honest that takes time.

Johanna Rothman describes the same thing from the buyer's side. You can look the information up anywhere, she says, but you won't get the value I add to it.

The pipeline problem the Susskinds missed

Drimalka points to something that turns the whole democratisation argument on its head. Managers, he says, are now turning to AI instead of junior staff for the work that used to be an apprenticeship. The entry-level tasks that built the next generation of experts are being automated away.

I watched a version of this happen once before. My mother was a typist from manual to electric. When computers arrived, the typing pool disappeared and every manager ended up doing their own typing. It felt like a gain, and in many ways it was. But the typing pool was also where a lot of people learned how an office actually worked.

Something similar is happening now. The person who uses AI to produce competent work today isn't necessarily building the pattern recognition that would let them work as a genuine expert in ten years. Procedure is getting cheaper and more accessible. Understanding is getting rarer. AI isn't closing the gap between the two. It is widening it.

What this means for experienced professionals

If you have spent decades building genuine expertise, the cheapening of procedure works in your favour. It clears the ground and makes it obvious what clients are actually paying for.

I now see the work I'm paid for sorting into two kinds. There is the procedure, increasingly cheap and increasingly AI-assisted. And there is the call about which procedure is the right one for this client, in this situation, at this moment. The second kind is scarce, and it can't be shortcut.

That is where your value sits. Knowing which of the steps is right, and when, and for whom.

Drimalka describes the professional who thrives as someone who combines life experience, domain expertise, and AI fluency. The AI does the processing. You decide which of its suggestions fits this client, in this situation, at this moment. AI amplifies expertise that already exists, provided you keep doing the thinking yourself.

The moment you let AI substitute for your thinking rather than inform it, you drift toward the procedural end of things, where your fees face downward pressure. So I use AI the way I did on that domain migration. It talks me through the steps. I watch, I check, and I decide. The tool does more of the work than it used to. The deciding is still mine.

This argument sits inside the wider Relational Cognition writing on expertise and how it compounds as procedure gets automated.

Frequently asked questions

References

Filip Drimalka on symbiosis with AI and the future of no work.

Wisepreneurs Podcast, on using AI as an extension of the mind and the narrowing pipeline for junior professionals.

Tomáš Baránek on deep reading as a professional edge.

Wisepreneurs Podcast, on why AI recombines rather than understands, and how ignorance raises the risk of being fooled.

Johanna Rothman on independent consulting principles that outlast every trend.

Wisepreneurs Podcast, on the value the expert adds to information anyone can now look up.

Richard Susskind and Daniel Susskind, The Future of the Professions (2015) — the argument that technology would democratise professional expertise by enabling less-qualified people to do specialist work

Every Tuesday I send The Wisepreneur, a short letter for experienced professionals building independent practice: one usable idea a week. Sign up at wisepreneurs.com.au/newsletter

When you want a second pair of eyes on your own practice, the Positioning and Practice Audit is how I work with people: wisepreneurs.com.au/marketing-partnership