At a Glance Career Capital is an Appreciating Asset
Three recent research papers from Anthropic, Stanford, and IESE Business School converge on a finding that runs counter to the dominant AI narrative: AI has made information cheaper, but the ability to transform information into genuine understanding, which develops through sustained hands-on practice, is becoming more valuable and structurally scarcer.
For experienced professionals, this changes the question from "how do I stay relevant?" to "how do I use what I've built?"
It also raises a question most AI commentary avoids: if AI is removing the pathways through which this capability develops, where does the next generation of experienced professionals come from?
The professionals gaining the most from AI aren't the most technically skilled. They're the ones who bring the richest distributed cognitive system, built through years of varied practice, to the collaboration.
Someone in your network refers a potential client your way, or they find you through your reputation and ask for a meeting. They want to explore whether you can help, and you want to get a sense of the real situation before committing to anything. They're not looking for a large firm with layers of account managers between them and the person doing the thinking.
Within the first few minutes, something registers. Not certainty, more like a quiet signal that what they're describing doesn't quite match what's actually going on.
Your hippocampus is already running pattern matches against hundreds of similar conversations you've had over a professional career, and it's flagging a gap between the stated problem and the one you're sensing underneath.
Twenty years ago, you'd have taken the brief at face value. Now, if you're paying attention to that internal signal, you notice the mismatch before your conscious analysis catches up.
That noticing capacity is part of a distributed cognitive system you've been building across your entire professional life, and it didn't build exclusively through client work. Your brain assembles prediction models from a much wider range of experience than most people realise:
- the case studies you've read and the papers that shifted your thinking,
- the stories colleagues told you about what went wrong on their projects,
- training where you practised under pressure, and
- life situations that taught you something transferable about how problems actually unfold.
A martial arts discipline like Krav Maga teaches you two things that map directly onto professional practice:
- when your first approach isn't working you need something else fast, and
- before you escalate you try to explore the situation through conversation and de-escalation to understand what the real problem is, if at all possible, because that's always preferable to a violent confrontation.
That same pattern, read the situation before committing, adapt quickly when your approach isn't working, and explore before escalating, is what experienced professionals do in discovery meetings, stakeholder negotiations, and crisis responses.
When a new situation triggers something that feels familiar, like a framework you have developed, yet something doesn't fit, your nervous system flags it. The quality of that signal depends on the breadth and variety of experience you've drawn from, not just the years on your professional CV, and it developed most powerfully in the moments your predictions were wrong and you had to figure out why.
What Anthropic found when they measured AI's real impact on professional work
Three recent research papers, from Anthropic, Stanford, and IESE Business School, have independently arrived at the same conclusion: what experienced professionals carry is not just knowledge or even judgment.
It is a distributed cognitive system, a combination of associative networks built through practice, professional relationships, tested frameworks, tools and environments you've learned to think with, and the embodied capacity to notice when something doesn't fit.
Cognitive science research by Anderson and Fast describes understanding as something that emerges from this kind of system, not from any single brain working alone. And that system is becoming structurally rarer.
Anderson and Fast's cognitive science research calls this constructing: the ability to synthesise genuine understanding from complex material.
AI is strong at finding information and moderate at filtering it. It remains weak at the kind of synthetic, judgment-dependent assembly of new understanding that experienced professionals do.
Information has become cheap.
Understanding remains expensive.
And the gap between the two is widening.
Not because experienced professionals are disappearing, but because AI is narrowing the entry points through which the next generation develops this kind of capability. Fewer graduates are getting the hands-on practice that builds it. Which raises a question most AI commentary ignores: where are the replacements coming from?
The finding that matters most for experienced professionals is who's actually affected. The workers in the most AI-exposed roles are educated, experienced, and earn significantly more than workers in less exposed roles, nearly four times more likely to hold graduate degrees.
This is about the kind of work that experienced professionals do, the knowledge-intensive, judgment-dependent work where your career capital either positions you well or leaves you exposed.
But here's what caught my attention. The researchers found no systematic increase in unemployment for these workers. What they did find was a 14% drop in job-finding rates for young workers entering AI-exposed fields.
Not mass layoffs of experienced professionals.
Fewer entry points for the people who would become experienced professionals.
Think about what that means over time. If fewer graduates are entering the fields where professional judgment develops through hands-on practice, the supply of that judgment shrinks. Not today. Over the next decade and beyond.
The researchers also found that what's actually happening with AI in workplaces is overwhelmingly augmentation: AI helping professionals do their work differently, not replacing them wholesale.
The automation that dominates headlines remains a fraction of actual use. Organisations are discovering, whether they articulate it this way or not, that professional work requires more than information processing. AI can search for information and filter what's relevant.
What it cannot do is the constructing work: synthesising genuine understanding from complex, contextual material where the situation evolves as you interact with it. That capacity belongs to the professionals who built it through practice.
The hidden cost of automating professional work
Two independent research teams, working separately at Stanford and IESE Business School, arrived at the same conclusion: automating professional tasks has a cost that standard productivity measures completely miss.
Enrique Ide at IESE built an economic model showing that when organisations automate entry-level work, they remove the apprenticeship pathway through which professional expertise has always developed.
The junior analyst who learned which questions to hold back, how to read a stakeholder's hesitation, when a data pattern is noise and when it matters, all by doing the groundwork alongside experienced colleagues, watching how they responded to situations, and gradually developing their own internal reference points? That pathway breaks when AI handles the foundational tasks.
His numbers are sobering. Even automating 5% of entry-level tasks can reduce long-run economic growth. A temporary 20-year disruption in knowledge transmission would be enough to erase the initial productivity gains from AI adoption.
And his model identifies something particularly relevant: even AI co-pilots that help rather than replace can hinder learning, because their reasoning is opaque.
Novices learn what to do but not why. They follow AI recommendations without developing the understanding that would let them exercise independent judgment.
Philip Trammell at Stanford found something equally unsettling at the individual workflow level. Tasks within a profession aren't separate units you can automate independently. They're connected through learning.
- Writing code teaches you to spot bugs
- Doing client research teaches you what questions actually matter
The foundational work is where professionals learn to search for what matters, filter noise from signal, and gradually develop the associative networks that eventually enable them to synthesise genuine understanding from complex material. Remove those tasks and you remove the learning, even when the direct output is preserved.
Researchers Shen and Tamkin made this concrete: using AI for coding tasks made programmers faster at individual tasks but worse at debugging and conceptual understanding. The speed gains masked a decline in overall professional capability.
Both research teams draw the same distinction: there's a fundamental difference between AI that helps you do better work (augmentation) and AI that does your work for you (automation).
Augmentation preserves and can extend your capability
Automation erodes the learning pathways that created it
This brings us to the uncomfortable question. Where are the replacements coming from?
If AI is narrowing the entry points through which young professionals develop expertise, and if even the "helpful" AI tools can undermine deep learning by making their reasoning opaque, the generation of professionals currently in their 50s, 60s, and 70s may represent an irreplaceable stock of accumulated judgment.
MacKenzie and Spinardi's research on nuclear weapons knowledge, cited in Ide's paper, offers a vivid precedent: losing just one generation of skilled engineers nearly caused nuclear weapons knowledge to be "disinvented.
Professional expertise is more fragile than it appears. It doesn't survive in manuals or databases. It survives in the people who developed it through practice.
What your brain has actually been building over the course of a career
Here's what the AI narrative consistently gets wrong about professional capability and ageing.
Neuroscientist Daniel Levitin's research on cognitive development shows that the kind of intelligence most relevant to professional work continues strengthening well into your 70s.
Researchers call it crystallised intelligence: the pattern recognition, the ability to read a specific situation and know which approach fits this organisation with these stakeholders at this moment, and the practical wisdom that builds through varied professional practice over the course of a career.
What makes crystallised intelligence so valuable is not just what you know. It's the distributed cognitive system you've assembled through sustained practice:
- the richness of your associative networks,
- the professional relationships you draw on,
- your practiced interactions with tools and environments, and
- the depth of mental models refined through every engagement where reality didn't match the plan.
Cognitive science research by Anderson and Fast describes this as a system where understanding emerges from the coordination of resources distributed across your brain, your body, your tools, and your professional relationships.
What declines with age is fluid intelligence:
- processing speed
- working memory
- the ability to solve entirely novel problems without relevant experience
These peak in your 20s and 30s.
But consider what professional work actually demands at a senior level. Not speed or the ability to hold more data in working memory.
What it demands is the capacity to generate useful predictions about how a situation will unfold, notice when those predictions don't match reality, and adapt your approach based on rich contextual understanding.
This is prediction work. Your brain has been building the architecture for it across every engagement, every difficult conversation, every project where the plan met reality and had to change. Your hippocampus matches new situations against patterns from hundreds of prior experiences, your nervous system registers familiar configurations before your analytical thinking catches up.
The learning didn't just happen when things went right. It happened especially when your predictions were wrong and you had to figure out why. Every time a consulting engagement that got hired bore limited resemblance to what actually unfolded, and you navigated the gap, your predictive models got better.
This capacity lives in your body as much as your conscious mind. It's embedded through years of varied practice and it operates faster than deliberate thought, which is why the signal arrives as a feeling before you can articulate the reason. The question is whether you're present enough to notice it.
AI research and neuroscience arrive at the same place: this kind of experienced and embedded capability is not just personally valuable. It's structurally essential. And the pathways for developing it in the next generation are narrowing.
The practical implication: staying cognitively sharp isn't about defending against decline. It's about maintaining the active engagement that keeps your prediction models updating. The moment you stop doing the cognitive work, you stop learning from the moments where reality contradicts your expectations. That's where your capability has always grown.
How this article was produced: AI partnership in practice
This is where I can speak from direct experience, because the article you're reading was produced through exactly this kind of partnership. Rather than describe the principle I can show you how it worked.
I brought three things to this collaboration with Claude:
- the Anthropic research I'd been analysing
- a study paper I'd already written synthesising the Stanford and IESE findings, and
- my years of accumulated notes from podcast interviews, client work, and research across cognitive science, complexity theory, and professional development
I store these in Obsidian, which Claude is able to access. Claude brought the ability to search across research sources and filter relevant findings, to cross-reference new material against what I'd already gathered, and to hold more threads simultaneously than I can track.
What Claude couldn't do was the constructing work: synthesising those threads into an argument that would resonate with how experienced professionals actually think about their practice. That required my associative networks, not its processing power.
The first plan Claude produced was too academic. It led with statistics and economic language: "61-percentage-point deployment gaps" and "strategic territory to occupy."
I read it and it didn't feel right. The language was too academic, too removed from how the work actually happens or even you would want to read. It needed to start with something a reader would recognise from their own experience, not with statistics.
Then the AI framed professional expertise as a kind of magical "sensing," and I had to redirect again, drawing on Jean Boulton's processual thinking and a recent conversation I'd had with a consulting colleague about how engagements actually unfold in practice.
Professional judgment works through prediction and mismatch detection, but the situations where it matters most are complex ones where there is no procedure to follow and no predetermined answer to apply.
This is part of why the client isn't talking to a Big 4 firm. Large consultancies bring systems, methodologies, and repeatable frameworks, and those work well for complicated problems with predictable components.
But the situations experienced independent professionals navigate are complex in a different sense: they're organic, they evolve as you interact with them, and the engagement that gets commissioned rarely resembles what actually unfolds.
Working in that kind of environment requires someone whose worldview treats the situation as processual rather than mechanical, someone who expects their initial reading to be incomplete and works with what emerges.
Your hippocampus matches new situations against patterns from prior experience, your nervous system flags what doesn't fit, and the learning happened in all the times those predictions were wrong and you had to adapt to what was actually in front of you rather than what you expected to find.
Then I asked the question that became the centre of the whole piece: where are the replacements coming from?
If AI is narrowing the entry points where the next generation develops expertise, what happens to the pipeline? That concern came from my realisation of how professional capability actually develops through practice. The AI wouldn't have generated that question without my lived context directing the conversation.
Each of those corrections came from my professional experience. The AI could process the research, but it couldn't tell when the output didn't read the way a practitioner would write it or when the framing missed the point.
This is augmentation in practice, and it goes deeper than the initial planning stage.
We (Claude and I) edited this article line by line, with me catching sentences that sounded like AI rather than a practitioner, replacing abstract language with specific professional situations, and redirecting the framing when it drifted into academic territory or missed the point.
Each correction drew on something the AI couldn't access on its own: whether a sentence would read the way a working professional actually thinks, whether the opening would feel real or manufactured, whether a concept like "contextual judgment" needed unpacking into what it actually means when you're sitting across from a client.
If you've built that kind of professional sensibility through varied practice, if you can tell when something reads right and when it doesn't, if you navigate situations that never unfold quite as expected, you already have what productive AI partnership requires.
The question isn't whether you can learn the technology. It's whether you bring what you've developed to the collaboration and stay engaged enough to keep sharpening it.
Where this leaves us
Three independent research teams, working from different angles, converge on the same place.
Anthropic's data shows that AI's actual impact on professional work remains far smaller than its theoretical capability. Stanford and IESE researchers show that closing this gap through automation rather than augmentation destroys the expertise organisations depend on, and narrows the pipeline that creates the next generation of experienced professionals.
And neuroscience confirms that the accumulated predictive capability of experienced professionals, the career capital built through varied practice, continues strengthening with age. Your brain hasn't been slowing down where professional work demands the most. It's been building richer, more nuanced models.
What the research points to, consistently across all three papers, is that the career capital experienced professionals carry, the distributed cognitive system assembled through varied practice, the practiced ability to construct genuine understanding from complex material, may be part of a shrinking stock of capability.
AI has made information cheaper. That has made the ability to transform information into understanding more valuable, not less.
Whether that applies to your particular expertise and practice is something only you can assess. But if you recognise what we've described in this article, the partnership model is worth exploring.
Career Capital is an Appreciating Asset FAQs
How do I know whether I'm using AI as a thinking partner or just handing work over?
Notice whether you're thinking harder or less during the process. If the AI is producing outputs while you watch, that's delegation. If you're directing the collaboration, evaluating outputs against your professional experience, and noticing where the AI's responses don't match what you know from practice, that's partnership. The distinction matters because one keeps you cognitively engaged and learning. The other lets your capabilities atrophy.
What about the younger professionals I work with or mentor?
This is where the research becomes urgent. If AI is narrowing the entry-level pathways through which professional judgment develops, experienced professionals who actively mentor and preserve hands-on learning opportunities become part of the solution. This isn't just professional responsibility. It's a way to increase the value of what you know, because the ability to transmit tacit knowledge becomes a premium capability when the traditional pathways for acquiring it are breaking down.
I can sense when something is wrong in a client situation but I can't always explain how. Is that real?
Yes, and it's not mystical. Your brain generates predictions based on accumulated experience. When a new situation doesn't match those predictions, you notice the mismatch before you can articulate what's causing it. This capacity develops through varied practice, including all the times your predictions were wrong. It's one of the capabilities that AI cannot replicate, because it requires embodied experience across contexts that never repeat in exactly the same way.
References
Anderson, S. P. & Fast, K. (2020). Figure It Out: Getting from Information to Understanding. Rosenfeld Media.
Automation, AI, and the Intergenerational Transmission of Knowledge, IESE Business School. Trammell (2026)
"Workflows and Automation," Stanford Digital Economy Lab. Shen & Tamkin (2026), How AI Impacts Skill Formation.
Levitin, D. J. (2020). Successful Aging: A Neuroscientist Explores the Power and Potential of Our Lives. Dutton.
MacKenzie & Spinardi (1995), Tacit Knowledge, Weapons Design, and the Uninvention of Nuclear Weapons.
Massenkoff & McCrory (2026), Labor market impacts of AI, Anthropic. Ide (2025)