Milos Mandic Can't Tell From 24 FDEs. Seven of Them Already Answered.

Milos Mandic published the State of FDE 2026 survey results yesterday, with 24 responses instead of the hundred he planned for, and ended on a question he says the data cannot answer. It can. Seven of his respondents answered it, and the artifact that reads their answer is already in The FDE Transition.

The short version

Milos Mandic's State of FDE 2026 survey on FDE Hub closed with 24 responses, and he held to the thresholds he set before opening it: no medians, no percentages, no compensation figures. The shape is still worth reading. Reporting lines split five ways, with three of 24 into product. Seventeen write production code for customers, and four said it merges back into the product. Only four expect the same work in two years. Seven of the 24 said the systems they build stay with them after handover, and twelve spend more than a tenth of their week keeping delivered systems running. Mandic offers two readings, a residency or a job that wears people out, and writes that he cannot tell which from 24 people. The seven can tell him. A residency ends with a handover, and in this series the handover is a keys line, a named person at the customer who can change the configuration without calling you. Seven people holding their own systems is the keys line blank at day 60, which is glue, and glue wears people out. His two closing questions are already two columns on the FDE scorecard, generalization rate and the keys line, and both are a product decision.

What 24 people said

Mandic's first move is the right one, and it is worth saying so before disagreeing with his last one. He wrote down his reporting thresholds before the survey opened, got 24 responses against a minimum of 100, and stuck to the thresholds. No percentages, because twelve people becoming "half of respondents" sounds like a market. No compensation figures, because seventeen people paid in dollars describe a room, not a rate. What 24 detailed answers can do, in his words, is roughly what 24 interviews can do: show the shape of something.

The shape. Twenty of 24 are in North America, 16 at AI product startups or scale-ups. Fifteen have been doing FDE-type work for under two years, and six of those fifteen have more than fifteen years of professional experience, so it is experienced people who moved into the seat recently. Twenty said the team grew in the last year. Nobody said it shrank.

The reporting line is flat across five answers. Six report to engineering, five to the CEO or founder, five to services or delivery, four to sales, three to product. In the open text, four people independently described the role as a product function. Seventeen write production code on customer engagements, and when he asked where it ends up, four said it gets merged into the core product, nine said it stays in customer-specific work, and the rest said both. Four of the seventeen described the loop Palantir built the role around. One respondent put the rest plainly: solutions are more custom than most would like to admit.

Then the cross that he calls the cleanest thing in the data. All four people whose week is 80% discovery write no production code. All four whose week is 80% building do. Two different jobs with the same title, which is the argument 2,000 of 17,000: The FDE Shortage Is a Product Problem made from the hiring side.

The question he leaves open

Fifteen of 24 think the role is becoming more critical as models improve. One thinks it is becoming less. At the same time, only four expect to be doing the same kind of work in two years. Seven expect to be leading an FDE team, five expect to be founding something, four expect a product or engineering role, three have no idea.

Mandic sits with that combination and offers two readings. One: this is a stage rather than a destination, a residency, and that is fine. Two: the job as currently set up wears people out. He notes that seven of the 24 said the systems they build stay with them after handover, and twelve spend more than a tenth of their week keeping delivered systems running, which he reads as some support for the second. Then: "I can't tell which one it is from 24 people." It is the question he most wants the next edition to answer.

He does not need the next edition. The seven already answered.

A residency ends with a handover

Month Two: The FDE Leaves, the Workflow Has to Stay is the page in this series about what happens at day 60, when the engineer's attention drops from one of three customers to one of twelve accounts held by a CSM who cannot read the configuration. The handoff is a page with four lines, and the one that matters here is the keys line: the named person at the customer who can change the customer layer without calling you. The rule on that page is blunt. If the keys line is blank at day 60, the deployment has not landed.

Seven of 24 saying the systems they build stay with them after handover is the keys line blank. Twelve spending more than a tenth of the week keeping delivered systems running is what a blank keys line costs, paid weekly, out of the discovery budget Mandic himself says it is coming from. Miku Jha at ServiceNow gave the failure its name, and The Career Out of the Seat: Draw the Exits, Not Just the Ladder carries it: the permanent glue between the customer and the product.

A residency has a handover built into it. That is what makes it a residency. Glue does not, and glue wears people out. The seven are not a stage. They are the second reading, and Mandic's own data says so without a single percentage.

The exits say the same thing from the other side. That post drew three doors out of the seat: into product, as the person who answers the handback packets you used to send; up, running the function without a bookings number; and out, to found a company on the problem you saw at three customers. Seven lead, five found, four into product or engineering. Sixteen of 24 named one of the three, counting the four headed for product or engineering as one door. People leave through doors somebody drew, and a seat whose code stays bespoke has no door into product, because there is nothing for product to receive.

His two questions are already columns

Mandic closes with two questions for anyone running an FDE team. Where does your FDEs' code actually end up? If the answer is always with the customer, you have a services team with a product team's title and cost structure. And who owns the things they shipped six months ago? If the answer is they do, you are spending your discovery budget on maintenance.

Both are right, and both are already on The FDE Scorecard: Count What Stayed, Not Who Was Sent. The first is generalization rate: the share of handback packets that got a written product decision within two weeks, and the share of three-customer patterns that shipped. Four of seventeen merging code back is that column, filled in by hand. The second is the keys line, which is what hours per deployment depends on; a function whose engineers still hold what they shipped cannot have that number falling.

Here is the part Mandic's reporting-line finding sets up. Both columns are a product team's decision. The written decision on a packet comes from product. The eval the generalized version has to pass comes from product. The handbook chapter Building the FDE Team: Reporting Line, Bar, Pairing, and Comp argues the line matters because it sets the scorecard, and that sales is the one line that turns the seat into something else within two quarters. In Mandic's room, three of 24 report into product and four into sales. He did not cross the reporting line with where the code goes, and this post does not either. But four people calling the role a product function in the open text, and three of 24 reporting there, is the gap between what the people in the seat think the job is and what the companies built.

One line on his hiring finding, because it matches. Two respondents, separately, put the ideal mix at 60% business and people skills and 40% technical, against an industry hiring for 80% technical. The handbook's bar is the product engineering bar plus the customer: production code, and the ability to learn an industry's vocabulary in weeks and earn an operator's trust. The people in the seat are saying the second half is being left off the job description.

Under AI Product Management the claim is that AI collapsed the cost of the parts PMs were trained for and left the parts nobody trained for. The deployment-to-product loop is one of those parts. Twenty-four people just described what it looks like when nobody owns it.

This week, if you run an FDE team: ask each engineer for the keys line on every deployment past day 60. A name, or blank. Count the blanks. That number is your answer to Mandic's question, and it is the one his next survey should ask.

Related answer: Is the forward deployed engineer role a residency or a job that burns people out?

Sources: Milos Mandic, "State of FDE 2026: The Survey Results," FDE Hub, October 27, 2026, read in full; every figure from the survey is as he reports it, and his earlier piece with John Gleeson on the role splitting is cited only as he refers to it. Falk Gottlob, "Month Two: The FDE Leaves, the Workflow Has to Stay" (October 6, 2026), "The Career Out of the Seat: Draw the Exits, Not Just the Ladder" (October 13, 2026), "The FDE Scorecard: Count What Stayed, Not Who Was Sent" (October 8, 2026), "2,000 of 17,000: The FDE Shortage Is a Product Problem" (September 22, 2026), and "Building the FDE Team: Reporting Line, Bar, Pairing, and Comp" in the handbook. Miku Jha's "permanent glue" line is cited as it appears in The Career Out of the Seat.