Roles dissolve when the constraints that bundled them go away. The sommelier rebundled around confidence under uncertainty. The radiologist around tumour-board judgement. The new bundle is the strategic move - not the new skill.
Every disrupted role has the same question: when the constraint that justified your work dissolves, what do you do?
The instinctive answer is to defend the existing role - more credentials, more certifications, more "AI-augmented" versions of the same work. The strategic answer is to rebundle around the next constraint, the one AI hasn't dissolved.
A role is a bundle of tasks held together by a constraint. When the constraint dissolves, the bundle dissolves too. The tasks may persist (typing didn't go away when word processors arrived), but they no longer cohere into a paid role with a career path.
Rebundling means identifying a different constraint - one that AI can't or won't dissolve - and building a new bundle of tasks around resolving it. The new bundle may look nothing like the old one. The sommelier's new bundle (curation, storytelling, decision confidence) was orthogonal to the old one (wine knowledge).
Rebundling is awkward at first. The new role doesn't have a name. Customers don't know to ask for it. Pricing is unclear. But over a 5-year horizon it produces durable advantage, where reskilling alone produces a treadmill that keeps accelerating.
A "dying" profession that became more central, not less.
In 2016, Geoffrey Hinton - the "godfather of AI" - predicted that radiologists would be replaced within a decade. Image interpretation is exactly what deep learning does. The prediction looked obvious.
AI does now outperform radiologists at many image-reading benchmarks. Demand for radiologists has grown anyway. Why? They rebundled. The job stopped being "interpret images" and became "integrate image findings with patient history, deliberate on tumour boards, hold accountability for treatment decisions in edge cases." All things AI can't do alone.
The radiologists who survived weren't the ones who became faster at image reading. They were the ones who rebundled their role around the constraints AI couldn't dissolve - multi-domain judgement, professional accountability, deliberation under uncertainty. The same playbook works across knowledge work.
Defensive rebundling buys time. Offensive rebundling builds the next position. Most professionals do the first because it's easier to organise around - and discover later that nothing was being built.
A small number of senior lawyers at large firms are rebundling around a new role: designing the AI systems that handle contract review, discovery, and routine drafting. They're not faster lawyers. They're the people who decide what the firm's AI does.
New title, new pricing model, much smaller cohort than the junior tier they're replacing. The defensive response was "more legal training." The offensive response is a different job entirely.
Old data team: analysts pulling reports, building dashboards, answering ad-hoc questions. AI absorbs most of that. The defensive response is "let analysts use AI to ship reports faster."
The offensive response: rebundle around designing the decision processes that AI now informs - what questions to ask, what data to ingest, what action to take. New role title nobody fully owns yet. Premium is forming.
The choice plays out over a decade. Walk the consequences.
Your role's old bundle is dissolving. You can rebundle defensively - keep doing what you did, lean on certifications and gatekeeping - or rebundle offensively, by claiming a new constraint AI hasn't dissolved. Different paths, different decades.
A job was a bundle of tasks held together by old constraints. AI dissolves the glue.
Learning new skills doesn't help when the system no longer needs human execution.
Roles thrive by resolving scarcity, risk, or coordination. AI moves which one is scarce.
Capabilities become rentable modules. The win is in the assembly.
When AI dissolves a bundle, value moves to whichever layer captures the new constraint. The map of who-wins-what gets redrawn.
The old bundle is gone either way. The question is whether you build the new one - or wait for someone else to.
An advisory engagement that redesigns the operating model and role architecture around the new constraints AI doesn't dissolve - building new bundles of work that hold their economic logic in an AI-restructured industry.