Exposure Isn’t Destiny: Rethinking Career Strategy in the Age of AI
Goldman Sachs put a number on it back in 2023: roughly 300 million jobs globally exposed to AI automation, with the technology eventually accounting for a quarter of all working hours in the US. Numbers like that travel faster than the caveats attached to them. On the newest episode of The International Risk Podcast, host Dominic Bowen sat down with Lucy Wark, a former McKinsey consultant, co-founder of career-skills startup Fuzzy, and the writer behind the Substack series How Not To Lose Your Job, to ask what those numbers actually mean, not for the labour market in the abstract, but for the person hearing them.
Why “Exposure” Doesn’t Mean What Most People Think It Means
Wark’s starting point is a distinction that gets flattened every time a headline compresses it into a single scary figure: in labour economics, “exposure” just means a job contains tasks AI can be applied to. It says nothing about whether that application replaces a worker or makes them more productive. Her framing is that exposure is a double-edged sword. When a new technology substitutes for a task, wages and employment in that field tend to fall. When it complements a task instead, making people more productive rather than replacing them, wages and employment tend to rise, and new roles get created alongside it. Most of the current debate, she argues, is really an argument about which of those two effects will dominate, not about whether AI is powerful.
The Five Pillars, and Why Adoption Is “a Puddle, Not an Ocean”
Wark breaks the topic into five separate questions that get routinely collapsed into one: capability (how good are the models, really), exposure (which tasks can they touch), adoption (are people and companies actually using them), economic impact (is it showing up in productivity and wages), and people impact (who is actually winning or losing right now). Her read is that the leap from exposure to impact is where most of the public pessimism outruns the evidence. Capability is rising fast. Exposure is broad. But adoption, in her words, is “a puddle rather than an ocean” — most users are still treating these tools like a chatbot rather than an autonomous agent, and at an economy-wide level, the effect on GDP, productivity and wages is barely detectable in the statistics so far. That tracks with McKinsey’s 2025 State of AI survey, which found 88% of organisations now use AI in at least one function, but only a third have moved past pilots, and most companies that do report an EBIT impact put it below 5%.
The Canaries in the Coal Mine
The one place Wark says the data is starting to move is at the bottom of the ladder. She points to Stanford’s “Canaries in the Coal Mine?” research, led by economist Erik Brynjolfsson, which used payroll data to track early-career workers in the most AI-exposed roles, software engineering and customer support among them, and found marked employment declines since ChatGPT’s release, concentrated specifically among junior staff doing the most process-driven, least judgment-heavy work. Alongside that, she flags a trend she calls “seniorization”: entry-level job postings that now quietly expect mid-level output, and early signs of contraction in freelance marketplaces for tasks like copywriting and design, work with essentially no employment protection standing between it and a cheaper AI substitute.
Which Career Strategies Just Expired
Wark wrote roughly 40,000 words on career strategy in 2024, before revisiting it from scratch in light of how fast the technology moved. Two of her old recommendations, she now says, no longer hold. Climbing the bottom rungs of a corporate ladder through loyalty to a single employer is a weaker bet if those bottom rungs are the ones narrowing. And credentialism, the assumption that a law or medical qualification guarantees demand for your labour, needs rethinking too, since so many of the tasks inside those professions are exactly the kind that can be automated, even as the credential itself remains a barrier to entry. She describes a “two-speed” effect already forming: junior professionals facing a narrower pipeline while senior practitioners in the same fields capture the productivity gains and see wages rise.
What’s Rising Instead: Platform, Proximity, Leverage
In place of the old playbook, Wark is now pointing people toward three strategies she thinks are strengthening rather than weakening. Platform building: putting your work, writing or voice in front of an audience directly, because a relationship with an audience isn’t something a model can substitute for. Proximity: investing in real human-to-human networks, digital and geographic, because owning relationships is more valuable the more replaceable individual tasks become. And leverage: applying the same effort to bigger pools of capital or bigger problems, since the tools now let small teams, or in some of the more extreme claims she’s seen floating around, a single person, operate at a scale that used to require dozens of employees. She’s skeptical of the most breathless versions of that last claim, but not of the underlying shift.
Even the AI Optimist Still Does It the Hard Way
Asked how she actually uses AI in her own work, Wark described a research routine that scans the literature on AI and labour markets daily and files it into a running knowledge base, work that used to take her hours of manual reading. But she was equally clear about where the tools stop being useful: every piece of writing still goes through three or four human editing passes, one for her own understanding, one for flow, one for voice, one for internal consistency, because AI can assist with each pass but can’t produce the finished judgment call. It’s the same instinct behind her worry about critical thinking in education: people retain a third of what they read, about half of what they discuss, and roughly 90% of what they have to teach. Outsourcing the struggle, she argues, is exactly how you lose the muscle that makes the struggle worth having.
The full conversation is available on The International Risk Podcast. For more of Lucy Wark’s writing on AI and labour markets, her series How Not To Lose Your Job is running on Substack.