Much of the debate on artificial intelligence is framed around a simple question: Which jobs will AI replace? This book asks a better one: Which parts of work become more valuable when AI makes many cognitive tasks cheaper?
What will still matter when intelligence becomes cheap?
Much of the debate on artificial intelligence is framed around a simple question: Which jobs will AI replace? This book asks a better one: Which parts of work become more valuable when AI makes many cognitive tasks cheaper?
The answer is not found only in technical skills, data processing, or routine analysis. As AI reduces the cost of producing text, code, summaries, classifications, predictions, and recommendations, value shifts toward what remains scarce: judgment, coordination, trust, accountability, tacit knowledge, and the ability to act in ambiguous human settings.
These are the “messy jobs.”
They are messy not because they are poorly organized, but because they involve problems that cannot be fully reduced to computation. They require holding coalitions together, navigating conflicting interests, interpreting incomplete information, persuading others, taking responsibility for decisions, and making change actually happen inside organizations.
This point is especially relevant for law.
AI can already support many legal tasks: preliminary research, document review, contract comparison, drafting, summarization, due diligence, and the identification of patterns across large bodies of material. These activities matter, and they will become faster and cheaper.
But legal work is not only the production of legal text or the retrieval of legal information. Lawyers and jurists operate in environments shaped by uncertainty, institutional context, strategic judgment, professional responsibility, ethical constraints, evidentiary limits, reputational risk, and trust. A legal answer may be formally plausible and still be strategically unwise, institutionally insensitive, procedurally risky, or professionally indefensible.
That is where the human role remains central.
The lawyer of the AI age will not be valuable merely because they know more rules than a machine. They will be valuable because they can decide which legal questions matter, assess which risks are acceptable, understand the client’s real interests, anticipate institutional reactions, negotiate among conflicting positions, and assume responsibility for advice in a concrete legal and social context.
The same is true for jurists more broadly. Legal reasoning is not just computation applied to rules. It involves interpretation, authority, legitimacy, prudence, and judgment under conditions of contestability. Law is a technical system, but also an institutional and political practice. That makes much of legal work a paradigmatic “messy job.”
The book’s central insight is economic: AI changes relative scarcity. When a capability becomes abundant, its price falls. When intelligence becomes cheaper, the premium moves toward the human capacities that complement it.
This means that AI will not simply replace or preserve jobs as they currently exist. It will unbundle and rebundle them. Some tasks will be automated. Some roles will shrink. Others will become more powerful because AI amplifies the people who can combine technical output with judgment, responsibility, and organizational skill.
For legal professions, this implies a profound shift. The premium will move away from routine production and toward legal strategy, institutional understanding, ethical accountability, client trust, interdisciplinary coordination, and the capacity to use AI without surrendering professional judgment to it.
That is why the future of work is also a future of organizations — including law firms, courts, public administrations, universities, companies, and international institutions. The key question will not be only what AI can do, but how institutions redesign roles, incentives, learning, authority, and trust around it.
Messy Jobs offers a rigorous and practical way to think about this transition. It moves beyond both techno-utopian and techno-dystopian narratives. AI will matter enormously, but human agency will not disappear. It will migrate toward the parts of work where context, responsibility, coordination, and judgment remain indispensable.
In an economy where artificial intelligence becomes cheap, the truly valuable work may be the work that remains irreducibly human.
Those who do research in a law faculty don’t disappear. But the source of their value changes.
The most fragile part of legal research is the part that resembles standardizable informational work: searching for sources, summarizing articles, reconstructing case law, producing literature reviews, comparing legal texts, preparing first drafts. These activities don’t become useless, but they become less distinctive, because AI drastically reduces their cost. This is exactly the logic of Messy Jobs: when certain forms of cognition become cheap, the economic and professional premium shifts toward judgment, coordination, responsibility, tacit knowledge, and human relationships.
The decisive point is that genuine legal research is not the same as legal information retrieval. A good researcher isn’t valuable simply because they “find” statutes, judgments, and doctrine. They’re valuable because they know how to formulate a relevant question, understand why that question matters now, situate it within an institutional system, distinguish a strong argument from a merely plausible one, assess the normative consequences of a thesis, and recognize the political or theoretical assumptions hidden behind a technical solution.
So the legal researcher of the future will be less of a compiler and more of an architect of the legal problem.
AI will be able to help a great deal with legal research. Cornell Law, for instance, describes AI as a tool that can assist in identifying patterns in legal texts, generating and testing legal theories, and expanding empirical legal study. But this is precisely what makes the human capacity to decide more important: which patterns matter, which hypotheses deserve trust, which correlations are legally significant, and which conclusions are institutionally sustainable.
For those working in a law faculty, the transformation will likely be asymmetric.
Those who produce purely descriptive research — “what does the literature say about X,” “what are the rulings on Y,” “how is Z regulated across five legal systems” — will be more exposed. That kind of work will be accelerated, automated, or devalued. It will remain useful, but it will no longer be enough to distinguish a researcher.
Those who can build an original contribution, on the other hand, will have more room. Original doesn’t necessarily mean “brilliant” or “revolutionary.” It means being able to do one of the things AI struggles to do reliably: redefine the problem, choose the theoretical framework, weigh competing principles, identify institutional implications, and connect positive law, theory, practice, and historical-political context.
In law, moreover, research is a typical “messy job” because it doesn’t work only with data. It works with authority, interpretation, legitimacy, conflict, hierarchies of sources, responsibility, institutional rhetoric, and practical consequences. A model can generate a formally elegant answer; the jurist must determine whether that answer is correct, defensible, coherent with the system, compatible with practice, and responsible in its consequences.
This is even more true for fields such as international law, EU law, constitutional law, administrative law, security law, criminal law, labor law, or the law of new technologies. In these fields, the legal problem is never purely technical. It is interwoven with sovereignty, institutions, power, fundamental rights, economic interests, security, democratic legitimacy, and conflicts between legal orders. AI can help map the material; it cannot replace the legal-political judgment that decides what that map means.
The most serious risk concerns the training of young researchers. If PhD students and research fellows use AI to skip the laborious stages of research, they may lose precisely the skills that build judgment: reading badly before reading well, getting hypotheses wrong, verifying sources, understanding why an argument doesn’t hold up, and learning the style of a disciplinary tradition. The UCL Laws document on the relationship between AI, training, and legal assessment insists precisely on this balance: AI can be a useful support, but the development of legal excellence remains crucial, and the risk of “cognitive offloading” can hinder learning.
So the question isn’t: “Will AI do legal research instead of us?” The question is: “Which part of legal research was actually research, and which part was just the cost of accessing information?”
The first part remains human and can become more valuable. The second becomes cheaper.