TL;DR

  • This is the sequel to What Does Expertise Mean When AI Can Pass Any Exam? - less about broken credentials, more about what expertise becomes next
  • Reference knowledge and routine pattern recognition are being commodified; judgement, accountability, integration, and tacit skill are appreciating
  • By 2030, “expert” likely means someone who can direct AI systems, bear professional liability for AI-augmented work, and teach skills that do not compress into training data
  • A concrete example: the 2030 civil engineer signs off on AI-generated structural calcs but remains expert at spotting when the model missed soil conditions the drawings never captured
  • The practitioners who win are the ones who classify their own work honestly and invest in the appreciating categories now

Expertise After AI argued that exams stopped measuring what we thought they measured. This post asks what replaces them - not as policy, but as a working picture of what practitioners will need to be good at by 2030.

What is being commodified

A few kinds of expert knowledge are being aggressively commodified by AI assistance:

Reference knowledge. Facts, procedures, standards, codes of practice. The ability to know the chemical interactions between two drugs, the legal precedent for a kind of contract, the building code requirement for a structural detail - all of this is increasingly available on demand from an AI that has the same access to the underlying references as the human expert does.

Pattern recognition in well-bounded domains. Reading a chest X-ray, recognising a malware signature, identifying the cause of a particular kind of failure. AI systems are reaching expert-level performance on many of these well-defined classification problems.

Routine application of established procedures. Standard medical diagnoses for common presentations, standard legal documents for routine transactions, standard engineering calculations for well-understood structures. The work that defined a large share of what professionals did is being automated.

Documentation and explanation. Producing clear written explanations of technical material, summarising complex documents, generating procedural instructions. Skills that were specialised and valuable are now available as a commodity.

The pattern is consistent: the kinds of expertise that consist of carrying information around in your head and applying it to common situations are exactly the kinds that AI systems handle well.

What is appreciating

At the same time, certain kinds of expertise are becoming more valuable, not less:

Judgement in novel situations. When a situation does not match any established pattern, when the available evidence is ambiguous, when reasonable experts could disagree about the right course of action - the human expert who can navigate this well is more valuable than ever. The AI can produce confident-sounding answers in these situations, but the answers are unreliable in ways that require expert judgement to detect.

The ability to ask the right questions. Working effectively with AI systems requires knowing what to ask, what to verify, when to push back, when to accept. This is itself a deep skill that depends on understanding the domain well enough to evaluate the AI’s answers critically.

Integrative work that crosses domains. Most interesting real problems span multiple specialised areas. The expert who can synthesise across them - the engineer who understands the business context, the doctor who can integrate the social factors, the lawyer who can think about the engineering of the system being regulated - is doing work that AI systems still struggle with.

Direct accountability. When the work matters enough that someone needs to be personally responsible for it - bear the legal liability, the reputational risk, the moral weight - the human expert provides something the AI cannot. The signature on the document, the seal on the design, the decision to admit the patient.

Tacit knowledge and apprenticed skill. The kinds of skill that cannot be written down because the practitioners themselves cannot fully articulate them - what Michael Polanyi called tacit knowledge, summed up in his line “we can know more than we can tell.” Surgical technique, courtroom instinct, the diagnostic sense that experienced clinicians develop. These are appreciating because they are exactly the kinds of skill that AI training cannot easily capture - if the experts cannot articulate them, neither can the text the models train on.

A concrete walkthrough: civil engineering in 2030

Take a licensed civil engineer reviewing a bridge retrofit in 2030.

The AI has already produced a compliant calculation package: load combinations per code, member checks, a draft drawing set, and a narrative justification that reads like a senior engineer wrote it. A junior engineer in 2020 would have spent weeks on the arithmetic. In 2030 that arithmetic is free.

What the signing engineer still does - and what the licence still means - is different work:

  1. Site judgement the model never saw. The borings showed acceptable bearing capacity on paper, but the engineer walked the abutment last month and knows the approach fill was placed in a wet season. The AI optimised for the soil report, not the mud line visible after the last flood.
  2. Stakeholder integration. The city wants minimal lane closures; the fabricator wants fewer unique plate sizes; the maintenance crew needs access hatches the cleanest design omits. The expert holds those constraints in tension - none appear as rows in the calc sheet.
  3. Accountability under uncertainty. The model recommends a retrofit scheme at 94% utilisation on the critical girder. The engineer accepts 88% because the corrosion pattern on the existing steel is worse than the photos in the training set suggested, and signs the drawing anyway.
  4. Teaching the next crew. The apprentice learns not by copying tables but by watching which AI outputs get rejected and why - tacit transfer that no prompt library captures.

The 2030 expert engineer is not slower at math. They are the person whose signature means the AI output was interrogated in the physical and political reality the project actually lives in.

What this might mean by 2030

If these trends continue for another four years, “expert” might come to mean something like:

A practitioner who can effectively direct AI systems in their domain. Not just use them - direct them, evaluate their output, catch their mistakes, push them toward the right answer. This is a much more active relationship with AI than the current “type the question, accept the answer” mode.

A practitioner with deep, specific, hard-to-replicate skill. Either in a domain that AI handles poorly (genuinely novel problems, tacit physical skills, interpersonal work) or at the cutting edge of a domain that AI handles for routine cases but not for the hardest ones.

A practitioner who can bear professional accountability for work that includes AI contributions. The 2030 expert engineer, doctor, lawyer is increasingly the person who signs off on AI-augmented work - and whose signature means something because they can detect the cases where the AI got it wrong.

A practitioner who can teach the work to the next generation. Apprenticeship survives where the skill is not captured in any AI. The 2030 expert is, in part, defined by the ability to transmit skill to others - which itself requires understanding the skill at a level deeper than performing it.

These four properties have something in common: they are all about the human’s relationship to the work, not just their performance of it. The expert of 2030 is not someone who can do what an AI can do, slightly faster. They are someone who can do the things AI cannot do, while also directing the AI effectively for the things it can.

What this asks of practitioners now

The implications for someone currently building expertise in a field are uncomfortable.

The traditional path to expertise involved years of carrying reference knowledge in working memory, building up routine pattern recognition, becoming faster and more accurate at the bread-and-butter work. That path still produces competent practitioners, but the value of what it produces is being eroded by automation faster than the building of it can keep up.

The new path involves building expertise in different things: judgement on hard cases, integration across domains, working effectively with AI as a tool, the kinds of tacit and interpersonal skills that AI cannot replicate. This is a harder path. It requires more deliberate choices about what to invest in. It does not have the same well-trodden curriculum that the traditional path has.

The good news is that the new path produces a more interesting kind of expert. The 2030 expert, if this speculation is right, spends less time doing routine work and more time doing work that is genuinely difficult and genuinely meaningful. The economic returns to expertise may be more concentrated at the top than they were, but the work being done at the top will be more interesting than the average expert work of the previous generation.

The honest summary

The meaning of expertise is changing, and the change is not symmetric across kinds of expert work. Some expert work is being absorbed by AI; some is appreciating in value; some new categories of expertise are emerging that did not exist before.

The practitioners who navigate this well are the ones who think clearly about which category their own work falls into, and who invest accordingly. The ones who keep doing what they have always done, on the assumption that the change does not really apply to them, are likely to find their position weaker than they expected when they look up in 2030.

The work of being expert in 2030 starts now. It is not a transition that announces itself - it is a slow shift in what makes expert work valuable, and the people who notice it early and adjust will be in different positions than the people who do not.