FDE Hub's Editor Took the Seat and Says Speed Moved the Scarce Part. His Survey Said Where: The Leaving.

Milos Mandic founded FDE Hub, the only publication dedicated to forward deployed engineering, and on November 10 he wrote that two weeks earlier he had taken the seat himself. This is a Field Note: what the field's own editor reports first from inside the job, and the answer to the question he leaves open, which he had already collected.

The short version

• Mandic joined Sycamore, which is building an agent operating system for the enterprise, because it is organised around the problems he had been writing about for a year.
• His first report has two sentences I cannot put down: "Speed like this changes which part of the job is scarce," and "I am still working out what that means in practice."
• Much of his day goes to directing agents, the pace astonished him, and the amount delivered is hard to compare with a year ago.
• His own State of FDE 2026 survey, 24 responses, already holds the answer as he reported it: seven said the systems they build stay with them after handover, and twelve spend more than a tenth of their week keeping delivered systems running. When the build is fast, the build is not the scarce part. The leaving is.
• What leaving takes, from The FDE Transition, is a page at day 60 with four lines and a recall rule, and going the other way a handback packet with product's written decision owed within two weeks. Agents can do most of the building now. None of them writes the keys line.

What he wrote

The post is short and personal, and it earns the title. He had spent a year arguing on FDE Hub that "agents earn autonomy in stages," that "trust decides how fast a deployment moves," and that there are "systems that go stale when the business changes its rules." Sycamore is organised around those problems, software for building AI agents and for governing them once they are running, so he went to work on them directly rather than describe them.

The seat he describes is the one this series has been describing. He sits in the engineering team of a very flat organisation. A large part of the job is research: "my job is to digest it, work out how it applies," and make it repeatable in the product. The other part is customers: "I sit with customers, work out what they really need," and bring it back. His line for the two halves together: "this loop is the reason FDE teams exist."

Then the field note. "The pace of the engineering has astonished me." "Much of my day goes into directing a variety of agents." The amount that gets delivered is hard to compare with his work a year earlier. And the two sentences the rest of this post is about: "Speed like this changes which part of the job is scarce. I am still working out what that means in practice."

FDE Hub stays vendor-neutral, the interviews continue, five new video conversations are recorded, and the essays will now draw on weekly customer work. It has, he says, been a good two weeks.

He already has the answer

In October I read his State of FDE 2026 survey, which closed with 24 responses, and wrote Milos Mandic Can't Tell From 24 FDEs. Seven of Them Already Answered. He held to the thresholds he set before opening it, no medians, no percentages, no compensation figures, and the shape was still worth reading. As he reported it: seventeen of the 24 write production code for customers, and four said it merges back into the product. Twelve spend more than a tenth of their week keeping delivered systems running. Seven said the systems they build stay with them after handover.

Seven of 24

Put that next to his two sentences. If the build is fast, and he says it is, then the build is not the scarce part of the job. What was scarce in his own survey, before the agents made the building fast, was the ability to leave. Seven of 24 could not. Twelve carried the upkeep. Speed makes that worse, not better, because an FDE who can deliver three times as much in a quarter can also accumulate three times as many systems that stay with them.

That is the answer to "which part is scarce." The leaving.

What leaving takes

The AI Product Management thesis is that AI collapsed the cost of the parts of the job people were trained for and left the parts nobody trained for. Nobody trains an engineer to leave. The series has two artifacts for it, and both are pages rather than meetings.

The first is the page at day 60, from Month Two: The FDE Leaves, the Workflow Has to Stay. Day 60 is when the FDE's attention drops from one of three customers to one of twelve accounts, held by someone who cannot read the configuration. Four lines. The promise, the handshake's three columns updated. The signal, the customer's own eval rows running on a schedule, with the go-live pass rate as the floor. The watch list, what this customer is likely to switch off, and what it did switch off. The keys, the named person at the customer who can change a prompt or a routing rule without calling you. Under it, a recall rule: what brings the FDE back, and how fast. A blank keys line at day 60 is the deployment not having landed, and seven of Mandic's 24 are a blank keys line that stayed blank.

The second is the handback, going the other way, which is the half of his loop that carries customer need into the product. Absorb Pain, Excrete Product: The FDE Handback makes it a packet: the delta between the shipped product and what the customer actually runs, in configuration terms; the customer's eval rows, the cases that failed on arrival and pass now; the reversals, what the customer turned off and why; and the count of other customers with the same problem. Delivered at two weeks, at go-live, and at day 60. Product owes a written decision within two weeks, from a named person: generalize, keep as configuration, or kill.

Neither page gets faster when the agents get faster. The eval rows still need the customer's experts to label them. The keys line still needs a person at the customer willing to hold the keys. The decision still needs a name on it. That is what scarce looks like from inside the seat: not the hours the build used to take, but the two pages that make the build somebody else's to run.

The note

Agents can do most of the building now. Mandic says so from two weeks inside, and I believe him. None of them writes the keys line, and none of them signs the generalize-or-kill decision. The scarce part of the job moved to the parts that were already scarce in his survey, and the survey has the column for it: how many of the systems you built this year are still yours.

This week

If you are an FDE: for every account you have shipped, write the keys line, a name, or the word blank. If you lead an FDE team: count the blanks. That number, divided by shipped, is the one Mandic is working out in practice, and it is the one I would ask him for in his next survey.

Related answer: What becomes scarce for a forward deployed engineer when agents do the building?

Sources: Milos Mandic, "To Write About the Work, I Have to Do the Work," FDE Hub, November 10, 2026, read in full; every statement about his two weeks and Sycamore is his. The survey figures (24, seven, twelve, seventeen, four) are his State of FDE 2026 results as he reported them and as I covered them in October. The one-of-three to one-of-twelve, the four lines, the recall rule, the packet, and the two-week decision are my own, from the series posts linked above.