The illusion of competence in the age of AI

The increasing use of AI tools in litigation has changed not only how juniors work, but how they learn. Research, first drafts and even structured argument can now be generated in minutes. The output is coherent, well organised and superficially persuasive. The risk is not that the work is obviously wrong, but that it looks “right.”

Litigation has traditionally been an apprenticeship. Junior lawyers learned slowly, through drafting, redrafting and receiving iterative feedback. That process was not efficient but it forced engagement with the underlying law. A skeleton argument, pleadings, witness statements did not emerge fully formed; they were built, tested and often dismantled along the way. Crucially, juniors learned why a point worked, not just how it should be expressed.

AI compresses that process. A well constructed prompt can generate a competent first draft in seconds. With further prompting, it can refine structure, tone and even introduce counterarguments. Prompting, itself, has quickly become a skill. But good prompting is not the same as good legal reasoning. It can produce a convincing answer without requiring the junior to test whether the answer is actually correct.

That creates an illusion of competence. The draft is polished, authorities appear plausible, structure is sound. What is missing though is the cognitive struggle that would ordinarily expose weaknesses. A junior who has not worked through the authorities is less likely to identify where the law is uncertain, where the facts do not fit, or where the argument invites an obvious response from the other side.

A junior produced a note on a limitation defence in a commercial dispute. The structure was clear and the conclusion confidently expressed. On review, however, the analysis relied on a line of authority that was distinguishable on the facts. That distinction was not obscure. It would likely have been identified had the junior engaged directly with the cases. Instead, the draft had been assembled through iterative prompts refining an initial AI output. The result was not obviously wrong, but it masked a fundamental gap in reasoning.

This matters more in litigation than in many other areas of practice. Litigation is not simply about stating correct legal propositions, but is about judgment. It requires an assessment of how arguments will be received, how they interact with the evidence and how they will be attacked. A technically accurate but poorly reasoned point can undermine credibility and weaken the client's position.

Clients are entitled to assume that advice rests on considered legal analysis. If AI-assisted work is adopted without sufficient scrutiny, there is a risk that advice appears authoritative but has not been properly tested. The exposure is not limited to obvious errors but to overconfident conclusions.

The courts have already shown a willingness to scrutinise the use of technology in legal practice. The recent decision involving Pinsent Masons (Cork v Smith), in which a junior relied on a fabricated statutory rule generated by AI, reflects a broader expectation that firms understand and supervise the tools they use. The analogy is straightforward: delegation to technology does not dilute responsibility, if anything, it increases the need for oversight.