Every organisational form of the last century has the same five parts: a strategic apex that decides, a middle line that supervises, an operating core that does the work, a technostructure that standardises it, and support staff that keeps the machine running. The shape exists for one reason: people who do the work have interests of their own and know things their principals do not. Managing that costs a firm most of what it spends on management.
Machine agents break both conditions. An agent has no career to protect, and everything it does can be logged at almost no cost. The apparatus built to manage human agency does not get cheaper. It becomes unnecessary. Which is why the transformation to the Principal Firm is not an addition to the organisation chart. It is a subtraction from it.
Three parts go. The operating core ceases to be human: drafting, analysis, first-pass production move to machine agents, and the work stays important while the firm stops holding it as a population of people. The middle line disappears. It existed to extend supervision across many hands, and there is nothing left to supervise when the record aggregates itself, continuous and unedited. The technostructure is absorbed into the specification. The standard is no longer written by an analyst beside the work but by the person who will answer for the outcome.
What remains is a single human part that specifies and answers, working above a specification, a population of agents, and a record that proves what happened. That is the whole form.
This is also why most AI programmes disappoint: they bolt agents onto all five parts and keep the apparatus, adding cost to a shape built for a different problem. The gains come from subtracting in the right order: groundwork first, then the operating model, then agents on the volume. The transformation is not installing AI into your firm. It is letting your firm take the shape AI makes possible.