A count of 5,426 job specs finds forward-deployed engineering hiding under other titles

A count of job specs is a thin instrument, and it is the only one in the record here. Chan describes the source as a recent report and does not name its publisher or the test it used for calling a role FDE-adjacent ^[16]. The two numbers he gives divide cleanly. Of 5,426 postings, 923 carry the FDE title, leaving 4,503 under other names, or 83 percent of the set ^{[1][2][17][18]}. Across 1,311 employers, that averages about four such roles per company ^[19].

Chan wrote that "the practice has gotten well ahead of the vocabulary" ^[21]. He liked the original case for these hires, and it was narrow: enterprise AI falls apart at the last mile, so send engineers past the last mile and cut the handoffs between sales, solutions and support ^[22]. He no longer thinks proximity buys much on its own. Research covering 1,500 FDEs put the median at around 47 percent of time working directly with customers ^[4]. That leaves 53 percent of the week elsewhere ^[20]. A vendor's promise that its engineers will sit with the customer's team, Chan wrote, amounts to the price of entry ^[23].

His incentive account does not depend on the count. Custom work is revenue already booked, and productizing it writes that revenue down in exchange for margin a company might see a year out ^[11]. Delivery leads have a utilization number to hit, and FDEs get promoted for shipping into named accounts ^[12]. Meanwhile the customer paying the bill wants their thing built now, and has no stake in a generalized version built later for somebody else ^[24]. "Nothing in that arrangement rewards the engineer who says we shouldn't build this a fourth time," Chan wrote ^[13].

Chan has a commercial interest in the category. He co-founded Second Talent, which places AI-native engineering talent across Asia and staffs teams in 10 Asian markets, so he sells the labour whose category he is naming ^[5][6]. That bears harder on the count than on the incentives. Readers can still check whether anyone inside their own company is paid for a deployment nobody needed. Chan says as much himself. "I can't see anyone's margin from where I sit. What I see is the hiring," he wrote ^[10].

Hiring orders are where this shows up. Delivery-first teams come back for roughly the same number of engineers with every new customer, while teams that push recurring problems into the product ask for a lot early, then fewer and more senior people as each deployment needs less custom engineering ^[9][8]. Where the learning stays in the heads of the engineers who did the deployments, Chan says, much of it leaves when they leave ^[7].

One change is cheap and available this quarter: a record, one sentence after each deployment describing what the customer could not do before the team arrived, kept separate from the technical scope ^[15]. The costly one lands in next year's headcount plan, when the customer-facing engineering line either tracks the account count or flattens. "Fix the comp plan and the calendar, and the behavior follows," Chan wrote ^[14].