The Disintermediation of Professional Services: What Agents Actually Replace, and What They Can't

Table of Contents

• What "Disintermediation" Really Means Here
• The Three Layers of a Professional Service
• Where Agents Are Already Cutting In
• Legal: The Associate Pyramid Is the Target
• Accounting and Audit: Reconciliation Goes First
• Marketing and Creative Agencies: The Execution Tier Collapses
• Management Consulting: The Deck Was Never the Value
• Why the Billable Hour Is the Real Casualty
• What Stays Human: Liability, Trust, and the Last Mile
• The New Firm Shape: From Talent Pool to Agent Operator
• Insights Most People Overlook
• References

What "Disintermediation" Really Means Here

Disintermediation is a word borrowed from finance, where it meant savers pulling money out of banks to lend it directly. The bank was the intermediary, and once depositors could reach borrowers another way, the bank's spread evaporated. The structure of the threat to professional services is identical. A client has a problem and an expert has an answer, and for a long time the only way to connect the two was to hire the firm, sit through the intake, wait for the associate to do the research, and pay for every hour in between.

Agentic AI-as-a-Service compresses that chain. When a competent vertical agent can take the same intake, run the same research, and return a draft that's 85 percent of the way there, the client no longer needs the intermediary for that portion of the work. They still might want a human to sign off. But they are no longer paying the firm to produce the answer, only to bless it. That distinction is where billions of dollars of margin live, and it is exactly the distinction agents are erasing.

The mistake most coverage makes is treating this as a story about job loss. It isn't, primarily. It's a story about which layer of a transaction captures the value. The same dynamic is reshaping software, where one agent can replace a ten-seat team and SaaS seat-pricing breaks; professional services are simply the labor-priced cousin of that shift.

The Three Layers of a Professional Service

It helps to break any professional engagement into three layers, because agents do not attack them evenly.

The information layer is research, retrieval, document review, drafting, and analysis, the raw production of work product. This is the most exposed. It is high-volume, pattern-heavy, and historically delegated to the most junior people in the building precisely because it's the most routinizable.

The judgment layer is interpretation: deciding which precedent actually controls, whether an aggressive tax position is worth the audit risk, what the data is really saying versus what it appears to say. Agents are getting better here fast, but judgment carries accountability, and accountability is sticky.

The relationship-and-liability layer is trust, advocacy, and the willingness to put one's name and license behind an outcome. A partner who tells a CEO "do not do this deal" is selling conviction and the credibility to be believed. No agent has a reputation to stake or a license to lose.

The firms in trouble are the ones whose pricing assumed clients would keep paying judgment-layer rates for information-layer work. That arbitrage, billing a $600/hour partner's rate structure on research a first-year does, was the quiet engine of professional services profitability. Agents kill the arbitrage, not the profession.

Where Agents Are Already Cutting In

Legal: The Associate Pyramid Is the Target

The law firm business model is a pyramid: a few partners leverage many associates, and the associates do document review, legal research, due diligence, and first drafts. The firm bills the client far more than it pays the associate, and that gap funds the partnership.

Agentic tools now do meaningful chunks of that tier. Contract review, e-discovery triage, clause comparison across a data room, and first-draft memos are increasingly handled by specialized legal agents that work overnight for the cost of inference. Thomson Reuters and the legal-tech wave it represents are explicitly aimed here; Gartner has projected that a large share of legal work will be touched by generative AI within a few years, and corporate legal departments are the ones pushing hardest because they pay the bills.

The disintermediation isn't that robots argue in court. It's that a general counsel can now get a defensible first-pass NDA review from an agent for a per-task fee instead of routing it to outside counsel at $450 an hour. The work that filled associate timesheets is exactly the work that's per-task priced now. The unbundling of legal services mirrors the great unbundling of SaaS suites by agents: clients buy the specific task, not the whole relationship.

Accounting and Audit: Reconciliation Goes First

Bookkeeping, reconciliation, transaction categorization, and variance analysis are textbook information-layer tasks, and they're being automated aggressively. The Big Four have all announced multibillion-dollar AI investments not out of enthusiasm but out of fear: a meaningful fraction of audit and tax-prep hours are mechanical, and mechanical hours are precisely what an agent does cheaply and tirelessly.

What survives, for now, is the attestation. An audit opinion is a liability instrument, the firm is staking its insurance and its reputation that the financials are fairly stated. An agent can do 90 percent of the testing and a partner still has to sign. But notice what happened: the firm's revenue from doing the testing collapsed toward inference cost, while the revenue from signing stayed. That's a far thinner, far less leveraged business than the one the Big Four built.

Marketing and Creative Agencies: The Execution Tier Collapses

Agencies are perhaps the most immediately exposed, because so much agency revenue is execution: producing variations of ad copy, building campaign assets, managing media buys, generating social content, doing keyword research. A client used to need an agency because they didn't have the people or tools to do this at volume. Now an agent does it at volume, on demand, per output.

The strategic layer, knowing which campaign to run, understanding a brand's positioning, reading a market, is stickier. But the dirty secret of the agency world is that much of the billed work was never strategy. It was bodies producing deliverables, marked up. Strip out the markup-on-execution and many mid-tier agencies have no business model left. This is the same squeeze hitting workflow-automation incumbents like Zapier in the agent era: when the doing gets automated, the middleman that organized the doing loses its cut.

Management Consulting: The Deck Was Never the Value

Consulting is the interesting case because consultants have always insisted the deliverable wasn't the point. They were half right. A McKinsey engagement produces analysis, benchmarking, and recommendations, much of which an agent can now assemble from public and licensed data in hours rather than the weeks a team of analysts once took. McKinsey's own research on generative AI's economic potential put the value of knowledge-work automation in the trillions, which is a striking thing for a consultancy to publish about its own labor input.

What consultants still sell that agents can't: the political cover of an outside name, the boardroom credibility to deliver bad news, and the accountability of a brand on the line. "McKinsey told us to" is a product. But the analyst-built model underneath it is increasingly an agent-built model, and that means the same conviction now rests on a fraction of the billable hours. The firm that figures out how to charge for conviction without the army of analysts wins; the one that keeps pricing the army loses.

Why the Billable Hour Is the Real Casualty

The deepest disruption isn't to any one profession, it's to the unit of account. Professional services are sold in time: hours, day rates, retainers that proxy for expected hours. The billable hour exists because, historically, expertise and time were inseparable. An expert's output scaled linearly with the hours they put in.

Agents break that link. An agent's output doesn't scale with time the way a human's does, which makes "how many hours did this take" a meaningless basis for pricing. The natural replacement is outcome or per-task pricing, pay for the reviewed contract, the reconciled ledger, the completed campaign, which is exactly the pricing model the GaaS category is built on, and which forces a wrenching reframe inside the buyer's organization from software budgets to labor budgets and back again.

This is why the billable hour is the casualty, not the lawyer. A firm can keep its lawyers and still die if it can't reprice. And repricing is brutally hard for partnerships whose entire compensation, promotion, and equity structure is built on hours. The incumbent's dilemma in software, cannibalize your own seats or watch a startup do it, has a precise professional-services analog: cannibalize your own billable hours or watch a leaner, agent-native competitor undercut you on the same outcome.

What Stays Human: Liability, Trust, and the Last Mile

It's worth being concrete about what doesn't disintermediate, because the surviving firms are organizing themselves entirely around it.

Licensed accountability. Someone has to be legally permitted, and legally on the hook, to sign the opinion, file the return, or appear in court. Agents can't hold a license or carry malpractice insurance. This is a regulatory moat as much as a technical one, and regulators are not in a hurry to dissolve it.

High-stakes judgment under ambiguity. When the facts are messy, the precedent is unsettled, and the cost of being wrong is enormous, clients want a human who will commit and be accountable for the call. Agents can inform this; they can't own it.

Trusted advisory relationships. The CFO who calls their banker for a gut check, the founder who trusts one specific lawyer, these relationships are built on years of reciprocity and read of character. They're slow to form and slow to erode.

The last mile of persuasion. Negotiation, advocacy, and getting a reluctant counterparty or board to actually move. This is performance and relationship, not information retrieval.

Everything around these cores, the research, the drafting, the analysis that feeds them, is fair game for agents. The defensible firm shrinks its headcount in the exposed layers and concentrates its humans where liability and trust live. That's a smaller, higher-margin, less leveraged firm. Which is why this is genuinely threatening to the industry even though "professionals" survive: the economics of the old pyramid don't.

The New Firm Shape: From Talent Pool to Agent Operator

The firm that thrives looks less like a talent pool and more like an agent operator with a thin layer of accountable humans on top. Concretely:

It runs a fleet of vertical agents tuned to its domain, contract review, audit testing, campaign production, and treats inference as a cost of goods sold rather than as a head to hire. It prices by outcome, not hour, which means it has to actually understand its unit economics, including the hidden cost of agent retries and the human intervention rate when an agent gets it wrong. It keeps a small bench of senior humans whose job is judgment, accountability, and relationship, the layers that don't disintermediate, and it pays them like the scarce assets they now are.

This is closer to a software company's margin structure than to a traditional partnership's, and that's the point. The professional services firm of the next decade competes on agent reliability and the quality of its accountable human layer, not on how many warm bodies it can throw at a data room. The intermediary doesn't vanish. It just stops being able to charge for the part a machine now does better.

Insights Most People Overlook

The arbitrage, not the labor, is what dies. Most analysis frames this as "AI replaces professionals." The sharper truth is that AI destroys the specific arbitrage of billing senior-rate prices for junior-level work. Firms that genuinely sell judgment were never running that arbitrage and are barely affected; firms that quietly depended on it have no model left, regardless of how many people they keep employing.

Regulation is the moat, and incumbents will weaponize it. The most durable defense isn't capability, it's the licensing and liability regime that says only a credentialed human can sign. Expect incumbent professions to lobby hard to expand what requires a licensed sign-off, not because the work needs it, but because it's the one moat agents can't cross. Watch bar associations and accounting boards, not just model benchmarks.

Disintermediation hits the training pipeline before it hits the partners. The associate and analyst tiers aren't just labor, they're how the profession trains its future judgment-layer experts. If agents do all the grunt work, where does the next generation of partners learn to exercise judgment? Firms that automate the bottom of the pyramid may be eating their own seed corn, and nobody has solved this.

Clients will under-buy accountability until it bites them. Because agents make the information layer so cheap, clients will be tempted to skip the human sign-off to save money, and most of the time they'll get away with it. The disintermediation will look complete right up until a high-profile agent-produced error (a bad filing, a missed liability) re-prices the value of accountable humans overnight. The market for the judgment layer will be volatile, not linearly declining.

The winning firms will sell their agents to their former clients. The endgame for a forward-leaning professional services firm isn't to hoard its agents, it's to productize them. When your competitive edge is a fleet of domain-tuned agents, the natural move is to sell access to those agents directly, becoming a GaaS vendor to the very clients you used to serve hourly. The line between "law firm" and "legal-agent platform" is going to blur from both directions.

References

• Gartner Newsroom, research and projections on generative AI in legal and corporate functions
• McKinsey & Company, The economic potential of generative AI: the next productivity frontier