Article 2.3 · The Human Skills Upgrade: Why AI Without Expertise is Just Expensive Autocomplete

T3PS Legal Dynamics · Series 2: AI Readiness — It’s Not a Technology Question ·

Written by Will Whawell. Human intelligence throughout; AI assisted with the drafting.

Refreshed June 2026

A few years ago, I was in a conversation with a partner at a small regional firm — the kind of conversation that has stayed with me. He wanted to talk about AI. Not about any particular use case, not about a specific problem he was trying to solve — just, broadly, about AI. He had heard it was important. He wanted some of it. When I asked what he was hoping it would do, the answer was approximately: "Help us be more efficient, like what the big firms are doing."

I have a name for this, borrowed from a friend who is an estate agent. The "my mate the estate agent" problem. Not thinking about value, purpose, or suitability — just wanting the thing that someone else has, because they seem to be doing well with it. It is not a uniquely legal phenomenon, but it is a particularly acute one in a profession where technology anxiety runs high and strategic clarity often runs low.

The point I want to make in this final article is one that is conspicuously absent from most of the AI-in-law conversation: the technology is the easy part. The limiting factor — in almost every case of AI underperformance or AI-assisted failure — is the human operating it. Or, more precisely, the human who lacks the skills to direct it, interrogate it, and know when to distrust it.

Prompt Engineering is a Genuine Skill

Let us be direct about something that is still treated, in some quarters, as slightly embarrassing to admit. Prompt engineering matters. How you instruct an AI system substantially determines what it produces. Context, precision, structure, the provision of relevant background information, the framing of the question — all of these affect output quality in ways that are significant and repeatable.

AltaClaro's analysis of AI skills for lawyers describes prompt engineering as requiring "a deep understanding of how the AI thinks, how it processes language, and how to guide it to produce useful, accurate, and context-aware results." For legal professionals, this translates into the ability to frame instructions that reflect the specific legal context, the applicable jurisdiction, the relevant facts, and the desired output format — all in a way that the AI system can process effectively.

This is not a low-level administrative skill. It requires domain expertise. You cannot write a good prompt for a litigation risk assessment if you do not understand litigation risk. You cannot effectively direct an AI to analyse a clause in a commercial lease if you do not understand what the clause is supposed to do and what risks attach to different formulations. The expertise is not replaced by the AI — it is the prerequisite for using the AI well. The humans directing the AI must be capable of framing precise, contextually rich instructions. Without that capability, the AI will produce fluent generalities, and the lawyer will have no reliable means of distinguishing the useful from the wrong.

Coursera's Prompt Engineering for Law specialisation and Spellbook's dedicated legal prompting courses represent an emerging infrastructure for developing this capability. That such courses now exist in dedicated legal formats is evidence that the profession is slowly beginning to take this seriously. It is not yet evidence that it is taking it seriously enough.

The Dictation Culture Problem

Prompt engineering, however, is only one dimension of the human skills problem. There is a broader cultural issue that predates AI and will survive it if not addressed: the legal profession's relationship with its own working practices.

Consider dictation. Many legal professionals — experienced, capable, productive by conventional measures — still work primarily through dictation. They produce letters, memos, and documents by speaking, having that speech transcribed or processed, and reviewing the result. There is nothing inherently wrong with this. It is a workflow that, for many people, is genuinely efficient.

AI could transform this. The same professionals generating two hundred pieces of dictation in a month, the majority of which are one-line confirmation emails and routine acknowledgements, could leverage AI to handle the standard communications entirely, freeing their dictation practice for the substantive work that genuinely benefits from it. The efficiency gain is obvious. The cultural resistance is also obvious, and it is not irrational — it is rooted in concern about quality, about responsibility, about professional identity. Managing that resistance is a change management challenge, not a technology challenge. And change management requires human skills that are not manufactured by an AI purchase order.

The Ethical Dimension

Smokeball's 2025 State of Law Report found that 53% of legal professionals express ethical concerns about AI — a figure that, reported accurately, should be read neither as evidence of Luddism nor as a problem to be overcome, but as a signal to be taken seriously. Ethical concern is appropriate. The fabricated citations in Ayinde v Haringey — where a barrister cited five AI-hallucinated cases — and the wasted costs order in Ndaryiyumvire for similar conduct are not anomalies. They are predictable outcomes when AI tools are used without adequate human oversight and critical evaluation.

The question is not whether to be ethically concerned. The question is whether the response to ethical concern is informed scepticism and rigorous oversight, or paralysis and avoidance. The firms producing the best outcomes with AI are not those that have eliminated ethical concern — they are those that have channelled it into governance structures, training programmes, and output review protocols. Ethical concern is the input. Rigorous practice is the output.

New Roles, New Capabilities

Emerging roles in the legal profession now include legal technology specialists, AI compliance officers, legal data analysts, and workflow optimisation managers. These are not theoretical future roles — they are being created now, by firms that have recognised that AI deployment requires dedicated human expertise rather than being absorbed as a side task by existing staff with other primary responsibilities.

The AI compliance officer role is particularly significant. As explored in Article 2.2, the COLP carries specific regulatory accountability for AI governance. But the COLP in many firms is already carrying a substantial compliance workload. The emergence of a dedicated AI compliance function — whether as a standalone role or as a clearly defined component of the legal ops or technology function — reflects the operational reality that AI governance cannot be managed informally alongside everything else.

Law schools are beginning to respond to this skills imperative. Programmes at Northwestern, Stanford, and Cornell, among others, are integrating AI literacy into foundational curricula — not as an elective enrichment but as a core competency. The Thomson Reuters Institute has called for a reimagining of how lawyers are trained, warning that AI risks eroding precisely the legal judgment skills that entry-level legal work has historically developed. For the profession as a whole, the upskilling challenge extends well beyond law schools and touches every point of the career ladder.

The Collaboration Deficit

There is one more dimension of the human skills problem that I want to address, because it is less commonly discussed and in my experience particularly stubborn: the legal profession's resistance to collaborative working.

Here is a practical example from costs work. A detailed assessment hearing approaches. There is a draft bill of costs, a points of dispute, and replies. The parties have legitimate disagreements about some items and are likely, if pressed, to agree on others. The sensible thing — the efficient, proportionate, professionally rational thing — would be for the costs lawyers on both sides to get on a video call, share a screen showing the budgets and the disputed items, and work through them methodically. An hour of structured negotiation could avoid half a day in the costs judge's list and save both clients meaningful sums in professional fees.

It almost never happens. Not because it is difficult — the technology is trivial, and AI tools could support real-time cost analysis during such a call with very little friction. It almost never happens because collaboration of this kind requires mutual trust, a shared interest in resolution, and the willingness to deprioritise procedural position-taking in favour of substantive problem-solving. Those are human skills. They are learnable. They are not currently being systematically developed.

AI amplifies whatever capability is directing it. In the hands of a skilled, collaborative, strategically thoughtful practitioner, it is a genuine force multiplier. In the hands of someone who lacks those qualities, it accelerates the production of poor work at a scale and with a confidence that poor work, historically, never had.

What the Skills Upgrade Actually Looks Like

The practical skills that need to be developed are not exotic. They are specific:

Data interpretation. The ability to read AI outputs critically — to identify where a system may be hallucinating, where it is producing plausible-sounding generalities rather than accurate specifics, and where human verification is non-negotiable. This requires understanding what the AI was trained on, what its known failure modes are, and what the consequences of a particular error would be.

Prompt crafting. As discussed: the ability to provide contextually rich, precisely framed instructions that direct AI systems effectively. This is domain expertise applied to a new interface. It requires practice, feedback, and ongoing refinement.

AI oversight and governance. The ability to design and operate review workflows that maintain professional accountability for AI-assisted outputs. Not reviewing everything twice in a way that eliminates the efficiency gain, but applying risk-proportionate oversight that preserves quality while capturing the benefit of AI-driven drafting and analysis.

Critical evaluation of outputs.Perhaps the most fundamental skill of all: knowing when the AI is right, when it is close enough, and when it is confidently wrong. This is not a skill that can be shortcut by better prompting. It requires genuine expertise in the subject matter.

The Prussian field marshal Helmuth von Moltke wrote in 1871 that "no plan of operations extends with certainty beyond the first encounter with the enemy's main strength." The same logic applies to AI deployment. The theoretical capability of a system counts for nothing if the humans operating it cannot adapt, interrogate, and override it when circumstances demand. The plan — in this case, the AI output — is not the result. It is a starting point. What happens next depends entirely on human judgment.

That judgment, in the legal profession, is the product of expertise, experience, and continuous learning. It is not rendered redundant by AI. It is made more important. Because the AI will produce outputs at scale, and someone has to be capable of distinguishing the excellent from the catastrophic.

AI only works well if people skills are upgraded. That is not a caveat. It is the central proposition.

Questions worth sitting with:

1.    In your firm, who has been specifically trained to critically evaluate AI outputs — not just to use AI tools, but to identify when those tools are producing plausible-sounding errors that only domain expertise would catch?

2.    Think about the last significant piece of legal work produced by your team. If AI had drafted the first version, what would the reviewing lawyer have needed to know to identify every material error? Does that reviewer currently have those skills?

3.    The firms most successfully adopting AI are investing simultaneously in technology, governance, and human capability — treating all three as interdependent. In your firm, which of those three is receiving the least investment, and what does that imbalance cost you?