AI commoditizes the residue of expertise

There is a line that has been getting passed around in AI labs for a while, and the more I sit with it, the more I think it is the right frame for what is about to happen to consulting fees. The line is that AI commoditizes the residue of human expertise, meaning whatever part of a discipline can be made explicit enough to train a model on. That is most of what consultants currently charge for. Not all of it, but most of it. And the part that is not in the residue is going to get more valuable, not less, while the rest collapses. Consulting fees are about to bifurcate sharply, and most consultants are still pricing as if both halves of the work are the same product. What “residue” actually means The framing is worth slowing down on, because the usual translation (“AI replaces consulting”) is wrong in a way that misses the actual mechanic. The residue of expertise is the part that has been written down, talked about, repeated in case studies, structured into frameworks, captured in playbooks. It is the documented surface of the discipline. Every consulting firm has a knowledge management system full of this material. Every senior consultant carries the same patterns in their head. The patterns are real and useful. They are also, by definition, the part of the practice that can be made explicit. What sits outside the residue is harder to describe. It includes judgment about which of fifteen possible frameworks applies to this specific client at this specific time. It includes the ability to sit in a board meeting and recognize, three minutes in, that the real disagreement is between two executives and the strategy question is a proxy. It includes the willingness to tell a CEO their last hire is the problem, when the consulting firm’s incentive is to keep the relationship warm. None of that has been written down. It cannot be made explicit without losing what makes it work. The models are extremely good at the first category and not good at all at the second. This is not a temporary limitation. The training data for the second category does not exist in a form a model can learn from. The judgment exists in the heads of practitioners who could not articulate it if they tried. The market has not adjusted prices yet. It will. The agency problem the models do not have A piece of the same line that does not get repeated as often is the part about agency. Current AI systems, even the most capable agents, lack something specific. They have autonomy for a given task. They do not have independent desires. Nobody woke up this morning, told Claude to design a marketing strategy, and had Claude decide instead to design a finance strategy because that was what the client actually needed. This sounds like a small distinction. It is not. The thing senior consultants do, the thing that justifies their fees, often involves disagreeing with the brief. The client says they want a market entry analysis. The senior consultant, three weeks in, says the client should not enter the market and should instead acquire one of the existing players. That redirect is the value. The brief was wrong. The consultant noticed. AI does not redirect briefs. It executes them. A junior consultant executes briefs too, which is why junior consulting work is the part of the practice that commoditizes first. The senior consultant’s job is structurally different, and the difference is the agency the model does not have. This is the part of the consulting practice that gets more expensive, not less, as the model capability improves. The contrast in price between a senior partner and a junior analyst is going to widen, possibly dramatically. Where the data already shows the shift There are some data points worth pulling out, because the abstract argument is easy to nod at and the concrete numbers make it harder to dismiss. One company recently disclosed that 65 percent of its support conversations now involve an embedded AI agent doing meaningful work in the loop. The remaining 35 percent are conversations the agent could not handle, escalated to humans. The humans in that 35 percent are not doing the same job they were doing two years ago. They are handling the residual hard cases, which means their average difficulty per case has gone up. Their pay has not adjusted yet. It will. The hard residue is where the cost concentrates when the easy work is automated. On GDPval, a benchmark that compares frontier models against human experts on economically valuable tasks, the current best models are hitting around 85 percent of expert performance. That number is not the whole story. The 15 percent gap is concentrated in tasks that require the kind of agency and judgment the models lack. The 85 percent the models can do is the residue. The 15 percent they cannot do is where the consulting fees survive. The mistake most consulting firms are making is treating the 85 percent as something they still need to staff for and charge for at consultant rates. It is not. The market price for that 85 percent is approaching zero. The fee survives only on the 15 percent. What this does to fee structure The standard consulting engagement is priced as a blended rate. Partner time, principal time, senior associate time, analyst time, all packaged together at a single hourly or project rate. The blending was always a fiction. It was useful because the client could not meaningfully tell the partner time apart from the analyst time, and the firm could subsidize the partner’s selling time with the analyst’s billable time. The blending is going to stop working. Clients can now run the analyst portion of the work in-house with an AI tool that costs a fraction of a junior consultant’s hourly rate. They will, increasingly. The portion of the engagement they will still pay for is the partner time, and they will pay for it at a higher rate than the historical blended rate implied, because they are no longer subsidizing the analyst time inside the same fee. The honest fee structure looks like this. Two tiers. Judgment work, priced at senior-only rates, billed in hours or as a project fee with no junior leverage. Deliverable work, automated, priced at near-zero or bundled in as included. The firms that move to this structure first will lose revenue on volume and gain margin on what survives. The firms that hold the blended rate will lose the business to the firms that unbundled, and to the clients who can now do the deliverable half in-house. The consultants who will do well The consultants who come out of this period in a better position are the ones who can articulate, for any given problem, what part of the work is residue and what part is judgment. They will charge differently for each. They will be honest with clients about which is which. They will not pretend the analyst work is hard when it is not. The consultants who come out of this in a worse position are the ones whose career was built on being faster, more diligent, or more thorough than their peers at the residue work. That work is no longer scarce. The thing that was scarce was the willingness to do the work at all. It is now not scarce. The model is willing to do it. This is the bifurcation. It is going to be visible in fee data within two to three years. Some firms will adjust. Most will hold the old blended rate until the market makes them drop it, which it will. The wrong response The wrong response to this argument, which I see a lot of consulting firms making, is to invest heavily in AI tools that make the residue work cheaper to produce internally. The reasoning is that if the residue is going to be commoditized, the firm should be the one commoditizing it, capturing the cost savings and keeping the blended fee structure intact. This will not work for very long. The clients have access to the same tools. The clients can run the residue work themselves. The arbitrage between “consulting firm with AI tooling” and “client with the same AI tooling” disappears within a couple of years, and the firm is left with no premium and the same eroding fee. The right response is the harder one. Accept that the residue is commodity. Charge accordingly. Build the practice around the judgment work that the models cannot do. Hire and develop the kind of consultant whose value is not in the deliverable but in the redirect, the disagreement with the brief, the willingness to say the uncomfortable thing. That is the consulting practice that survives. Most firms will not pivot to it because the existing fee structure is too lucrative to give up voluntarily. The market will give them no choice within five years. The firms that move first will look smart in retrospect. The firms that wait will look like the law firms in 2010 that thought document review was going to stay a billable category forever. AI commoditizes the residue of expertise was originally published in DataDrivenInvestor on Medium, where people are continuing the conversation by highlighting and responding to this story.