The Principal Problem: Why Consultants Must Reclaim Epistemic Authority in the Age of AI

We have not yet applied it to ourselves in this age of supermaxxed AI.

When a consultant delegates analysis to an AI system, a quiet inversion occurs. The model generates a line of reasoning. The consultant reviews it, adjusts the language, and presents the output as professional judgment. On the surface, this looks like the principal directing the agent. But look more carefully. Who shaped the structure of the argument? Who determined which evidence was salient? Who decided what counted as a conclusion? In most cases, the model. The consultant has become the agent of the model’s outputs, not the sovereign author of their own reasoning. The client – the actual principal in this relationship – receives analysis whose epistemic origins are obscured even from the person presenting it.

This is not a technology problem. It is a first principles problem. And the first principles in question are very old.

The mediaeval European university organised its foundational curriculum around the Trivium: grammar, logic, and rhetoric as the core technologies of mind, the toolkit by which a thinker could receive a claim, interrogate its structure, and communicate a position with integrity. Grammar taught you to read precisely. Logic taught you to reason without flaw. Rhetoric taught you to persuade without deceiving. Together, they constituted what we might now call epistemic sovereignty: the capacity to know how you know what you know, and to be accountable for it.

Supermaxxed AI – LLMs on speed, really – has made the Trivium urgent again.

Large language models are extraordinary at one thing: producing text that resembles the outputs of reasoning. They are trained on the (largely stolen) outputs of human creativity, and they reproduce its surface patterns with astonishing fidelity. But resemblance is not equivalence. A model does not evaluate the validity of an inference; it predicts what a valid-sounding inference looks like. It does not weigh evidence; it generates the kind of sentence that typically follows a body of evidence. This distinction – between the form of reasoning and its substance – is precisely what the Trivium was designed to illuminate. It is also what years of professional practice, at its best, develops in a consultant.

The risk is not that AI will replace consultants. The risk is that consultants will allow AI to replace the part of themselves that makes them worth engaging in the first place.

I have seen this in practice. An engagement team produces a rapid diagnostic using AI-assisted analysis. The frameworks are coherent. The language is assured. The slides are clean. And yet, when the client asks a probing question about the underlying assumptions, the room goes quiet in a way it shouldn’t. The team had reviewed the outputs. They had not reasoned through them. The difference is invisible until the moment it isn’t.

Reclaiming epistemic authority does not mean refusing useful tools – no one ever accused consultants of being stupidly self-sabotaging. It means using them as a consultant uses any instrument: subordinate to judgment, not substituted for it. Concretely, this requires three habits that mirror the Trivium closely enough to be uncomfortable.

First, read the claim, not just the output. Before accepting an LLM-generated analysis, identify its central assertion and ask whether you could reconstruct the argument independently. If you cannot, you do not yet own the reasoning.

Second, interrogate the inference. Where does the logic move from evidence to conclusion? Is that move valid, or merely plausible-sounding? LLM systems are particularly prone to generating confident conclusions from ambiguous evidence. Old-fashioned logic catches this where familiarity with the prose will not.

Third, take rhetorical responsibility. When you present a position to a client, you are not a conduit. You are an author. The reasoning in the room carries your name, so act accordingly.

The Trivium was not designed for an age of artificial intelligence. But it was designed for exactly this problem: the problem of a thinker who has access to more received knowledge than they can critically process, and who must decide, under conditions of uncertainty, what actually holds up. That is the consulting condition in any era. It is simply more acute now.

We advise clients to return to first principles when complexity overwhelms convention. It is reasonable to apply the same prescription to ourselves. The principal in the room should always be the human. The question is whether we are prepared to put in the work, ironically, to be one.

Jules Yim is a Senior Consultant at The Cynefin Company and works at the intersection of complexity, organisational foresight, and human-AI collaboration globally. She is also a co-founding curator of Seapunk Studios, a research-and-imagination lab building brighter solarpunk futures for her native Southeast Asia.