AI is making your job description less useful

Job descriptions were always a rough approximation of a job. AI is widening the gap between the label and the actual surface area of the work — which changes what a manager is for as much as it changes what one person can do.

Job descriptions have always been a rough approximation of a job. They list the responsibilities, the reporting line, the skills somebody thought mattered when the role was written, and rarely describe how good people actually work once they’re inside an organisation. AI is widening that gap considerably, because it gives an individual access to capabilities that used to sit elsewhere in the business, take much longer to acquire, or require somebody else’s help — researching a market, interrogating a pile of documents, prototyping an idea, structuring a project, synthesising customer feedback, building a lightweight tool, all without waiting to be formally assigned the work.

Your job description might say content writer. The practical surface area of what you can now investigate and contribute to could stretch into customer research, product marketing, distribution, operations and product development.

That sounds like an uncomplicated productivity win. It isn’t, because the same tools that make it easier to become more useful also make it easier to spend a week doing sophisticated work that nobody particularly needed.

Capability isn’t the same as priority

Before generative AI, understanding an unfamiliar market or working through a large pile of documents took enough manual effort that the cost itself forced a decision about whether the task was worth it. AI removes a lot of that friction, and removing it removes the discipline that came free with it.

Ask one question and you get another avenue to investigate; feed in some documents and another pattern emerges; spot a problem and within minutes you’re designing a solution. Every path feels productive because something tangible keeps appearing on the screen — which is exactly what makes it dangerous.

A well-prompted afternoon disappearing into competitor analysis and speculative product ideas can feel considerably more serious than procrastinating on social media. It can still be procrastination.

That’s the individual, felt version of an argument I’ve made at the level of hiring trends in the editorial jobs are moving upstream: as production gets cheap, the scarce thing stops being whether you can do something and starts being whether you have the discipline to decide that you should. The question your job description was never going to answer is which of the things AI now makes possible for you actually deserve your attention.

The job description is a boundary, not a full instruction

The obvious response — write much broader job descriptions — could easily make things worse. Own content, customer insight, AI experimentation, distribution, product feedback, analytics and strategic innovation isn’t a progressive job description. It’s seven jobs stapled together.

The better interpretation is that a job description should work as a boundary and a starting point rather than a complete instruction set, which is probably how the strongest people in any organisation already operated before AI existed. They understand what they’re responsible for, but they notice things around the edges: they learn enough about adjacent disciplines to have better conversations, they spot gaps, they connect work happening in different teams, they solve small problems rather than automatically routing them onward.

AI dramatically increases how far one person can operate that way. Someone working on an article can investigate the search environment, analyse previous company content, compare the narrative with competitors, and look at what customers are asking, before producing a first serious draft — without that meaning they now own SEO, competitive intelligence and distribution. It means they understand enough of those things to make a better editorial decision.

AI can expand your awareness without your responsibilities needing to expand at the same rate, and that distinction is doing most of the work in this piece.

This makes your manager more important, not less

There’s an argument that AI reduces the need for management, because employees can increasingly answer questions themselves. I think that confuses two different things. AI can reduce your dependence on a manager for information — what has the organisation already published about this, what research exists, what does the market look like.

It can’t reliably tell you what actually matters inside the organisation right now. It doesn’t know that a project which looks strategically important on paper has quietly lost executive support, or why one stakeholder relationship needs more care than another, or which apparently small piece of work matters enormously because of what it connects to. That’s organisational context, and giving it to you is what a good manager is actually for.

So the relationship changes shape rather than shrinking. The weak version is tell me what to do and how to do it. The stronger, AI-era version is closer to help me understand what matters and where the boundaries are, then let me work out the route myself.

That needs trust running both ways: enough autonomy from the manager that exploration doesn’t get treated as going off-script, and enough discipline from the employee to know when to come back to the actual job rather than let exploration become the job.

AI can help you navigate the organisation, not just the internet

One of the more useful things AI does isn’t producing work faster. It’s helping you understand where you actually are. Large organisations are difficult to navigate — strategies, decks, research, old campaigns, policies and years of decisions scattered across different systems and other people’s memories — and you normally learn that gradually and imperfectly, by asking colleagues and finding an old presentation.

Given appropriate access and governance, you should increasingly be able to ask the organisation itself what it already knows about a customer problem, where a topic has appeared before, which claims get made repeatedly and what evidence actually supports them.

That’s the same underlying move I’ve written about for a personal knowledge system in context is capital and GitHub for knowledge — durable context that AI can be pointed at, rather than knowledge trapped in one person’s head or one platform’s history — just applied to an employer’s institutional memory instead of your own. It makes you less dependent on organisational folklore, and less dependent on your manager being the router for every document, which frees them to spend more time on judgement and priorities instead of forwarding things.

You still need rules for stopping yourself

There’s a less glamorous skill underneath all of this: stopping. The more capable AI becomes, the more you need rules that stop exploration from consuming execution, which in practice means knowing, before you start, what the actual outcome is, who it’s for, what would count as enough research, what doesn’t belong in this piece of work, and when you stop exploring and ship.

Without those questions, an AI-assisted worker can become extremely good at intellectual tourism — impressive notes on interesting problems, and nothing that actually changes. The expensive thing stops being the work itself and becomes choosing the wrong work to spend a week on.

(If your honest position is that you’d rather not be doing any of this exploring in the first place, minimum viable AI is the companion piece — the same discipline, aimed at somebody who wants the smallest defensible amount of AI in their job rather than the largest.)

The same problem is sharper for freelancers

A freelancer or one-person business experiences an extreme version of this, because there’s no department around them to absorb the slack — no research team, no sales function, no product manager reminding them what needs doing next. AI can fill pieces of all of those roles: researching markets, analysing competitors, drafting proposals, building simple tools. For somebody building a business alone, that’s extraordinary leverage.

It’s also a trap, because almost everything suddenly looks defensible as “working on the business” — rebuild the website, start a newsletter, research five customer segments, write a twenty-page strategy document — while nobody has actually bought anything.

For a freelancer, AI therefore raises the value of having a genuinely simple answer to who you’re trying to help, what problem you’re solving, what evidence you have that they care, and what action is most likely to move you closer to revenue this month. AI can help you execute almost any direction competently. That’s exactly why choosing a direction becomes the harder and more important decision.

Learning without waiting for permission — with a real complication

Expanding into an unfamiliar discipline used to mean finding a course, getting training budget, or hoping somebody experienced had time to explain it. AI adds a route that doesn’t require any of that: you can build a highly specific learning process around the exact problem in front of you, ask it to teach you the concepts you need, interrogate terminology you don’t understand, and arrive at a conversation with a specialist better prepared than you’d otherwise have been. That reduces a kind of professional helplessness — “I don’t know anything about that” becomes less acceptable when the basic knowledge is one conversation away.

I don’t think that’s the whole story, though. I’ve written elsewhere about the uncomfortable flip side: in raising the floor is not building a ladder, my own project evidence found that AI-assisted work helped the least experienced person the most — producing a credible result sooner than they’d otherwise have managed, which is a claim about output, not about whether the judgement behind it actually formed.

On-demand learning driven by curiosity, the kind this section is describing, is a genuinely different situation from being handed a finished draft to sign off — but I’d be overclaiming to say AI straightforwardly solves professional helplessness when I’ve also argued it can just as easily let competence arrive ahead of comprehension. Both are plausible. I don’t think I know yet which one dominates.

For the same reason, I’m not going to rebuild the seniority argument here either — raising the floor and career leverage, not the ladder already make the case, with actual evidence, that accumulated experience is turning into judgement rather than speed as the thing that distinguishes a senior person from a junior one with good tools. It’s the same distinction this piece keeps landing on: AI can produce the analysis. It can’t decide whether the analysis is a distraction, whether the evidence is weak, or whether a technically good piece of work is organisationally useless.

What this asks of an organisation

Some organisations say they want people to use AI while still routing every decision through the same hierarchy that predates it, which creates a straightforward tension: give people tools that make them more capable of investigation, then only let them perform tightly prescribed tasks, and a lot of that capability goes unused. The opposite extreme — everybody should investigate and innovate constantly — isn’t better; it’s a recipe for ten people quietly solving the same problem.

What actually helps is more ordinary than either: real clarity about which problems matter most, where people are genuinely encouraged to experiment, which decisions need specialist or managerial input regardless of what AI can produce, and how a useful discovery gets captured rather than lost in one person’s chat history.

I’ve written about the employer’s side of that bargain already — the short version is that “just use AI more” is not a policy, and organisations that give people the capability without the clarity are the ones who’ll be surprised by where a week actually went.

The scarce skill is direction

Capability is becoming abundant enough that this is turning into the recurring theme of everything I’m writing about AI at work right now — I made the same point about model access specifically in when the best AI model stops being worth it: capability is abundant, context and judgement aren’t.

The job version of that argument is the same shape. If execution keeps getting cheaper, more things become possible; when more things become possible, deciding what actually matters gets harder, not easier, and that’s true whether you work inside a large company or for yourself.

AI gives you extraordinary freedom to move beyond the narrowest reading of your role. That doesn’t mean the role should become unlimited — it means you need a much clearer sense of what the job is actually there to achieve.

The job description becomes the outer label. The manager supplies context and direction. AI expands what you’re capable of doing inside that space. And your judgement decides whether any of it was worth doing at all.