AI is compressing legal work but increasing the value of divergence

Much of the discussion around AI in professional services still focuses on capability, i.e., how powerful are the models? How quickly will work change? Which firms are 'ahead'? These are valid questions, but they may not be the most important ones. The deeper issue is what happens when everyone begins using the same systems, trained on the same patterns, producing the same 'credible' outputs.

Within the legal and risk profession today, AI hasn't just created efficiency but also convergence. And that convergence creates risk.

I found myself thinking about this while stuck in traffic recently. Google Maps suggested a side street to avoid the congestion ahead. Fair enough (helpful even), except several cars in front of me were receiving exactly the same recommendation and peeling off into the exact same side street. The algorithm was not wrong in identifying an alternative route, but the new problem was that once everyone followed the recommendation, the alternative path simply became another traffic jam.

This feels increasingly relevant to how organisations are beginning to use AI.

The rise of synthetic consensus

One of AI's most interesting characteristics is its ability to (quickly) produce outputs that feel highly polished, highly plausible and highly aligned to accepted thinking. That sounds useful and often it is, but there is an emerging executive risk hiding inside that capability: synthetic consensus. When leadership teams, consultants, lawyers and advisors increasingly rely on similar models trained on similar information patterns, organisations risk converging toward the same assumptions, the same recommendations and ultimately the very same strategic decisions.

Importantly, this convergence may not feel obvious because the outputs themselves are usually coherent and can even feel more polished than traditional human-generated work. The danger is not that the work looks wrong but rather that it looks convincingly right.

Historically, organisations benefited from friction. Different advisors approached problems differently, and junior teams researched issues from first principles. Partners brought personal judgement shaped by sector experience, relationships and instinct. Disagreement was inefficient, but it also created some diversity of thought. AI compresses much of that variation.

If every firm asks similar questions of similar systems, then the resulting advice may begin clustering around a narrower band of acceptable thinking, and that is strategically dangerous territory. 

The familiar return to old law firm economics

Interestingly, AI may also be forcing law firms back toward something much older…

For the last century, large law firms have become increasingly industrialised, and scale has come through leverage. Work was broken into stages, and time became the overriding commercial unit. The billable hour emerged because legal services increasingly resembled structured production systems.

But that was not always our model. Historically, clients often paid for judgement, relationships, trust and outcomes. Senior lawyers were valued less for visible 'production' and more for discernment, strategic navigation and contextual understanding. 

In many ways, AI may be reversing the economics that created the industrialised law firm. If drafting, summarisation, research and even certain forms of analysis become increasingly automated, then the value of simply producing legal work begins to decline, and what becomes scarce again is judgement. Not generic answers or polished outputs.  The ability to identify which path among several plausible paths is actually the correct one is a very different commercial proposition.

It also explains why conversations around value pricing, relationship pricing and charging for strategic insight are accelerating again across the profession. 

Businesses are about to discover which processes actually matter

There is another implication sitting underneath all of this, too.

For years, organisations accumulated layers of process because labour was fragmented, coordination was difficult and human time was expensive. Entire workflows emerged simply to move information between people and AI has exposed this very quickly. When systems can draft, route, summarise and synthesise information instantly, many organisations will discover that parts of their process architecture never created meaningful value in the first place. Some processes exist because they genuinely improve judgement, governance or outcomes, but others simply exist to manage the limitations of human coordination.

AI acts as a form of operational truth-telling and this may become uncomfortable because once friction disappears, organisations are forced to confront which parts of their workflow were creating strategic value and which parts were merely administrative choreography.

Competitive advantage may become divergence

When every system suggests the same side street, somebody still needs the judgement to decide whether following the crowd is actually the right move.

For years, firms competed on scale, efficiency and consistency but AI will make those capabilities increasingly accessible to everyone, meaning competitive advantage may increasingly come from differentiated judgement. The firms that succeed may not be the firms generating the fastest answers, but the firms best able to resist synthetic consensus long enough to arrive at better ones. 

So what have I decided to do about this? Increasingly, the discipline is not validating the most plausible answer, but interrogating it: if these are the most plausible answers, what is the most implausible question nobody else is asking? In an environment optimised for coherence, the real risk may not be bad answers, but simply 'believable sameness'.