The generalist's comeback

For most of my career, being interested in everything made it harder to fit into one job. AI hasn't made me an expert in everything. It has made it easier to do something useful with the connections I see.

Two things are happening at once. AI is making people like me more capable than we have ever been at the work itself. And the systems that decide who gets hired are getting narrower. Both of those are AI. Only one of them is on my side.

The older you get, the more you are expected to be one thing

The older you get, the more the world expects you to have become an expert in something.

At the beginning of a career, curiosity looks like potential. Nobody asks a 26-year-old to explain the thread running through their CV, because there isn't supposed to be one yet. Then the questions change. What is your specialism? What are you senior in? What have you owned? Where do you see yourself in five years, and why doesn't your CV already show it?

None of these is a hostile question. They are the ordinary vocabulary of hiring. But they all assume the same shape: that experience should narrow over time, that a career is a line rather than a set of overlapping circles, and that by a certain age you should be able to name the one thing you are.

I've never had a clean answer. I've spent more than 20 years across marketing, technology and startups, in different industries and different kinds of companies, moving towards whatever interested me or whatever problem seemed worth solving. I learnt a lot about a lot of things. I never became the recognised expert in one field.

When I was younger, that felt like something I still had time to fix. As I got older, it started to feel like something held against me. Seniority gets measured in years inside one discipline: the title you can point to, the team you have grown, the depth you can defend in an interview. By that measure, breadth looks like a lack of commitment.

And yet the breadth was useful. I could usually understand enough of a problem to see how it connected to something else: that what looked like a marketing problem was really an operations problem, or that a repetitive task could be solved with a tool the team wasn't using. Friends came to me for the same reason. At work, I was drawn to the things that sat outside the neat boundaries of my role.

That could be valuable. It could also be a problem. A company does not hire you to be broadly interesting. It hires you to do a particular job, and the more senior you become, the more clearly you are expected to represent one discipline.

Part of that is how my attention works, which I wrote about in the last post. But I also used to explain the rest as too many possibilities. There is truth in that: other people's problems come with boundaries, and your own life doesn't. But paralysis is only part of it. The deeper problem was expecting that, with enough time, breadth would turn into depth in one recognised field. My career did not follow that shape.

Both things can be true: I had something useful to offer, and I did not always fit the structure I was trying to offer it in. The easier you are to categorise, the easier you are to hire. I became harder to categorise at exactly the age when I was supposed to be easier.

AI fills the gaps

It fills in the parts I am not good at.

Most jobs are not one skill. They are a stack of small ones: having the idea, writing it up, building the spreadsheet, sending the email, remembering to do it on Thursday. If one of those is where you fall down, the whole thing falls down with you, and it looks like you are bad at the job when you are only bad at a part of it.

AI can cover that part. Not perfectly, and not without me checking it, but well enough that the idea gets out of my head and becomes something real.

That is not a superpower. It is closer to being handed the step everyone else was already standing on. And it matters well beyond generalists. If reading and writing is slow for you, or keeping things organised is the hard part, or the language you work in is not your first, or your energy has to be rationed across a day, the missing piece was never the thinking. It was the step.

And there is a reason it works well for a generalist. Knowing a little about a lot means you usually know which question to ask, and how to ask it. Ask a vague question and you get a confident, useless answer. Ask a precise one and you get most of the way there. And when the answer arrives, you have enough context to notice that it is wrong.

There is a second half to this, and it is quieter. The connections I see between marketing, operations and technology used to be worth something only if I could get them built. Now I can build a rough version myself and find out whether the idea was any good.

Which is why "the comeback" is the right word and "superpower" is not. Nothing was added to me. Something that was always there became usable.

The same technology, on both sides of the door

The hiring process has been quietly automating, and it is not built to recognise this kind of person.

Most large companies don't read CVs first any more. Applications arrive in an applicant tracking system, an ATS: software that parses them into fields, filters and ranks them. If you have applied for a job online in the last ten years, your CV has already been through one.

The narrowing doesn't need a villain. It happens through standardisation: a parser reads what fits its template, a recruiter works down a ranked list because there are 400 applicants and a finite afternoon, and a job description is written in one discipline's words.

The evidence that this happens before a person is involved is hard to argue with. According to Greenhouse's Future of Hiring Candidate Report 2026, four in five job seekers across the US, UK, Ireland, Australia and Germany were rejected in the past year without speaking to a single person at the company (HRD). One 2026 survey of employers found that 65% said AI rejects applicants before a human reviews them (MyPerfectResume).

The direction is clear. The first reader of my CV is increasingly a system scoring it against a description of a role that was written for a specialist.

This is where the irony sits. The tools that make a generalist more useful are the same tools that make it easier to filter generalists out. Both are doing exactly what they were asked to do. One is helping me cross the gaps between fields. The other is being asked to find the person who already lives in the middle of one.

My problem is the small version of a bigger one

My case is the mild version of a bigger problem.

Some of it is mechanical. A CV with two columns, a table, a header, an unusual file name someone chose in a hurry: any of these can make a parser read the wrong thing, or nothing at all. People get filtered out for how their CV was built, not for what's in it.

Some of it is statistical. A model learns from the people you hired before, and then prefers more of them. Amazon built a recruiting tool trained on ten years of applications, most of them from men, and it learned to penalise CVs that mentioned women. They scrapped it (BBC).

Race, age, nationality, a gap in a CV, a name: the model does not need to be told to prefer one kind of person. It only has to be trained on who kept getting through.

None of this means the people who built these systems wanted it. The intention was reasonable: 400 applications, one afternoon, find the plausible ones faster. But a tool that sorts people inherits every assumption in the sorting, then applies them to everyone, at scale.

That is the real difference from a biased human. A person can be surprised by you. A person can read something odd in a CV and think, hang on, there's something here. A ranking cannot, and increasingly there is nobody sitting next to it to have that thought.

Rules are starting to appear. New York has required annual bias audits of these tools since 2023 (NY State Comptroller), and the EU AI Act treats hiring software as high-risk. An audit is a floor, not a fix.

And the more applications are written with AI, the harder the filters squeeze. Which pushes out precisely the people whose value was never in the keywords.

So we end up with a strange split. In the work itself, breadth is becoming more valuable. At the gate, it is becoming less legible. The comeback, if it comes, may arrive for the people who no longer need to pass the gate.

The comeback has limits

I like the phrase "the generalist's comeback", but I don't mean that specialists are becoming irrelevant.

The opposite may be true. The easier it becomes to produce a plausible first answer, the more important real expertise becomes when the consequences matter. Someone still needs to know when the model is wrong, when the shortcut is unsafe and when a problem is more complicated than it looks.

AI can help a generalist reach further. It cannot remove the need for judgement, verification or depth. And it creates a new danger: knowing just enough to produce something convincing without knowing enough to see what is missing.

Part of the work is knowing when you are not the right person, and saying so.

A different shape of work

I still have too many interests, and I can still disappear down a rabbit hole.

But I am beginning to think the problem was not being a generalist. The problem was repeatedly trying to fit a generalist's way of thinking into roles built around a narrower kind of progression.

AI doesn't fix that on its own. It gives people like me more ways to turn breadth into visible work: something to build, test, show and improve. If you are a generalist reading this, I don't think the answer is to wait for the gate to open. The evidence you need is a thing that works, not a better word for what you are.

That is roughly where I ended up. Not with a better CV, but with a different container for the work. I stopped trying to get through the door and built pleko instead. How that happened, and why this shape fits what I am actually good at, is the next post.

The generalist's comeback may already be happening. It just isn't happening at the door.