Start With the Box You Already Have

Every team I have worked on, at every company and every size, has had some version of the same small group at its center. Three people, give or take, who together decide and make the thing: someone who owns the problem and the what, someone who owns the how it is built, someone who owns the how it is experienced. The names change, product and engineering and design at one company, product and engineering and architecture at another, but the shape is so common that people stopped seeing it as a choice and started treating it as the definition of a team. Call it the box, the three-in-a-box, because that is what most of the industry calls it and because nearly everyone reading this has sat in one. It is the unit of execution that a generation of software was built by, and the reason to start a part about staffing the agentic team there is that it is the model already in your head, and the honest way to describe what the agentic product needs is not to replace that model but to show you where it has to grow.

The word grow needs one guard against a misreading, because the previous parts of this book have argued hard that an agentic product needs roles the box never had, the architect who enforces the boundary, the eval owner who keeps the checkmark honest, the supervisor who watches the running agent. It would be easy to read that as a call to blow up the box and draw a new org chart with eight or ten seats. That is the wrong reading. The box is not wrong. It is the right starting point at any size, because it encodes something true, that a small number of people holding the decision, the build, and the experience, in tight collaboration, is how things get made. What the agentic product does is not invalidate the box. It loads work onto the box that the three seats cannot all hold, and the question this part answers is which new seats the box grows, in what order, at what scale, and the answer starts from the box rather than from a list of titles.

The box was always growing

The first thing to notice, and the thing that makes the growth feel natural rather than disruptive, is that the box has been growing the whole time. It was not always three. Earlier it was two, a product lead and an engineering lead splitting the what and the how, the pattern people called two-in-a-box. Design earned a seat and it became three, because as products became things people experienced rather than merely used, the experience needed an owner at the table and not a service called in at the end. At some companies a fourth seat was added, a business owner, or a data owner, or an architect, because the work demanded a kind of judgment the existing three could not supply. The box has a history, and the history is a sequence of the same move: the work changed, a kind of judgment became load-bearing that used to be occasional, and the box added a seat to hold it. Nobody experienced those additions as the destruction of the team. They experienced them as the team maturing.

So the agentic shift is not the first time the box has had to grow, and framing it as the next instance of an old and successful move is both more accurate and more useful than framing it as a revolution. The agentic product changed the work, the way every prior expansion was triggered by the work changing. It made a kind of judgment load-bearing that used to be occasional, several kinds, in fact, and the box has to add seats to hold them, the same way it added design and then business. The difference, and it is a real difference this part will get to, is that the agentic shift does not ask for one more seat. It asks for several, and they cluster in a part of the work the box never had to staff at all, the supervision of a thing that acts on its own. But the move is the familiar one. You are not throwing away the box. You are doing to it what every successful team has done to it before, growing it to fit the work, and the only thing new is how much it has to grow and where.

Growth is not status, it is work

There is a reason to insist on the box-grows framing beyond its accuracy. The previous part named the conversation that keeps the supervision column empty, the contest over who is central, and the grid is the antidote because it grounds every seat in work rather than rank. Staff from the work and a new seat lands as relief, someone finally owns the thing that was falling through; staff from the org chart and the same seat lands as threat.

What the rest of this part does

So the structure of the answer is set, and it is worth laying out plainly before the chapters that fill it in. The grid from the previous part, the cells of work across the two channels, is the map of what has to be owned. The box, the three-in-a-box you already run, is where the ownership starts. And staffing the agentic team is the act of growing the box to cover the grid, which means deciding, for a team of your size, which cells the existing three seats can still hold and which cells need a new seat, a new hat, a new owner, because the work in them has become too much or too different for the box as it stands.

The chapters ahead do that in two moves. The first is the team member the box did not know it added, the agent itself, which turns out to be staff and not software, and which lands hardest on the one seat in the box already living the supervisory life this whole book has been describing. The second is the staffing itself, how the grid gets covered at a startup of five and an enterprise of five hundred, which seats merge and which must separate, and which of the new owners the agentic product needs are roles the industry has not yet named or learned to pay for. Through all of it the box stays the unit, because the box is what teams have and what teams understand, and the agentic team is not a replacement for it. It is the box, grown to fit a kind of product the box was never asked to make before, and the growing is the work of this part.