The gap is a redesign gap

The high performers in the survey did one thing differently. About three quarters of them redesigned workflows around AI. About a quarter of everyone else did.

That is the whole story of the missing EBIT. Productivity gains land on the individual. Profit lands on the process. If the process does not change, faster people produce the same output in less time and the company pays for the same headcount.

Large firms are scaling agents anyway. The share running agents in at least one function went from 27% to 40% in a year. That is a lot of agents deployed inside processes nobody redesigned.

Walid Mehanna, chief AI officer at Merck KGaA, put the durable edge in four words in July: context, data, processes, and people's fluency. His second point is the one boards should read twice. Leaders now have to run a workforce that is part people, part agents, and most of them have never done it.

That is the work. Redesign the workflow, then lead the hybrid team that runs it. It is also, almost exactly, what a senior fractional operator sells.

What got cheaper inside McKinsey

McKinsey published its own numbers in January and restated them in August. Roughly 25,000 agents working alongside roughly 40,000 people. 1.5 million hours saved, mostly in search and synthesis. Client-facing roles up about 25%, non-client-facing roles down about 25%, output up about 10%.

Read the mix, not the headline. The hours that disappeared were the hours spent finding and summarising. The roles that grew were the ones sitting across from a client.

Synthesis is now cheap. A junior analyst's week of desk research is an afternoon for an agent. Big Consulting is publicly worried about the billable hour because the hour it used to bill was full of exactly that work.

What did not get cheaper: knowing which of three plausible recommendations survives contact with this board, this CFO, this regulator. Knowing when the model is confidently wrong. Signing your name under the decision.

A fractional who sells synthesis is competing with an agent on price. A fractional who sells the call is competing with nobody in the room.

The next buyer is an agent

This is where we put our own stake in the ground.

Today an expert is hired by a person. Someone writes the brief, takes three meetings, and signs a statement of work. The hour is the unit, and the RFP is the channel.

Our view at Bridge AI is that a second channel is forming underneath that one. Companies are putting agents inside their processes, the McKinsey numbers above show how fast, and an agent running a workflow will hit a point where it stops being confident. At that point it needs a human, in-session, for the length of the exception, at a price set in the moment. The protocols to make that call already exist; the plumbing is being built now.

In that channel the agent hires the expert. The expert never sees the ninety percent of the workflow the agent handled alone. They see the ten percent that needed a human, and they are paid for that ten percent.

We have not seen this at scale yet. What we have seen, across every intake call we run, is that the matching problem is already moving from people scanning profiles to systems consuming evidence. The agent as buyer is where that line ends.

What a fractional sells to an agent

An agent calling for help does not read a CV. It matches on four things, and most fractionals have never written any of them down.

Context. Which industries, which company sizes, which situations, stated precisely enough that a router can match on them. "Ten years in finance" matches nothing. "Carve-out working-capital resets in PE-backed industrials, 50 to 500 million revenue" matches.

Trust. Evidence that a past call was right, with the outcome attached. A structured record of decisions and what happened after them, in a form a machine can read.

Liability. Someone has to own the answer. The agent cannot. The platform will not. The expert who signs is the product.

Exception handling. A clear description of where the automated process breaks and what the human does when it does. That is the job the agent is hiring for, so it goes first on the page.

The practical move for a fractional this quarter is to build that evidence base. A profile written for a recruiter reads well and matches poorly. Structured skill rows, bounded claims, and logged outcomes match well and read fine. This is the reason we turn every intake call into skill rows before we do anything else.

Converting personal leverage into retainers

Most fractionals have already picked up personal AI leverage. They draft faster, model faster, summarise faster. The trap is passing that leverage to the client as a discount.

The McKinsey survey gives the better path. Boards cannot book AI EBIT without a workflow redesign and someone to lead the hybrid team. That is a retainer with a business case attached, and the fractional's own AI fluency is the credential for it. The gain from your tools funds the depth of your engagement, not a lower day rate.

Two honest caveats. The survey is a survey; 37% attributing EBIT impact tells you about attribution, not about ground-truth returns. And the agent channel is our thesis, with early plumbing and no audited volume behind it. The RFP still pays this year's invoices.

Still, the direction is set. Search and synthesis have already been repriced. The next thing to be repriced is the handoff, and the fractionals who have made themselves legible to the thing doing the handing off will be the ones it calls.

Sources