A discipline, not a toolkit.

The documentation captures what the expert knows they know. The real value is in what they do not know they know.

In any long-running business, there is a layer of judgment that never makes it into a procedure. The pause before responding to a difficult client. The mental discount rate applied to a quote on a Tuesday versus a Friday. The subtle pattern that says this deal is going to close and that one is not.

This is the tacit layer. It is the reason a firm with the same website, the same SOPs, and the same staff as a competitor can produce twice the revenue. It is also the reason transmission fails. When the owner steps back, the tacit layer walks with them.

The industry response to this problem used to be documentation. Write it all down. It does not work. Tacit knowledge is tacit precisely because the expert cannot articulate it.

Transfer of Experience is the formal discipline for this problem. It has existed for decades inside regulated industries where moving a production line between countries requires a team to deliver at spec on day one, and where documentation alone has never been enough.

The craft has a shape. You embed alongside the expert. You shadow work. You note what they do, especially the things the SOP does not mention. You build a second operator who can run the job the way they do. You stress-test it. You iterate. You hand over.

Before AI agents, the endpoint of that work was a set of revised procedures and a trained team. It took months of human embedding and years of calibration. Expensive, slow, but reliable when done seriously.

Today, the endpoint is something else. A custom agent that has absorbed the expert's voice, rules, and corrections, running alongside the team, drifting in the direction the expert would drift. The craft did not change. The medium did.

Two things changed in the last eighteen months.

Agents can hold voice. Current models can write in the owner's voice with enough fidelity that clients do not notice. That used to be impossible and is now a solved problem.

Agents can hold rules. Structured prompts, memory files, and tool use let us encode the owner's judgment rules explicitly. When the owner says we never send a proposal on a Friday, the agent can be made to respect that without retraining, without fine-tuning, without magic.

That combination is new. It means the old Transfer of Experience discipline can now produce something owners never had before: a running system that keeps doing the work when they step back, in their voice, under their rules, measurable and correctable.

Most engagements begin with the Audit. Two to four weeks, fixed price. Small team interviews, shadow sessions, workflow maps. At the end, you have a document. A diagnostic, not a pitch. One that names what is in your head and what it would take to extract it.

If the diagnostic is right, Build starts. Eight to sixteen weeks. The agent goes live on one operator first, not the whole team. That operator becomes the tuning partner for the first month. We capture the corrections they make. Those corrections become training data.

Operate runs indefinitely after that, on retainer. We do not step away. Most agents need quarterly retuning as the business evolves, and every three months we add scope. Another workflow, another signal, another use case. Clients who try to run Operate themselves get agents that degrade in six months.

Transmit and Scale are optional. Most owners enter them twelve to twenty-four months after Build, once they trust the agent enough to consider handing it off.