Why your autonomy reqs sit open 90 days.

Every autonomy and ADAS hiring manager I talk to has the same req. Perception engineer, sensor fusion, motion planning, controls. Opened a quarter ago. A trickle of agency resumes that miss the mark, two onsites that went nowhere, and a VP asking why the program milestone is now at risk.

The req is not cursed. It is mispositioned. I have spent 15 years recruiting in automotive and mobility, including a stretch as the only recruiter for the U.S. launch of AFEELA at Sony Honda Mobility, where ADAS and autonomy engineers were one of nine functional categories I hired across. The same five problems stall these reqs everywhere, and every one of them is fixable without adding headcount or agencies.

First, look at what candidates see

I run a free job board that tracks open roles across EV, AV, and autonomous delivery companies, pulled nightly from their own career pages. Here is the snapshot from this morning:

Two things in that data should change how you run your req.

You have more direct competitors than you think. Nuro alone lists 79 open roles right now. Scout Motors and Wayve are visibly scaling too. Your perception req is not competing against the company across town. It is competing against every funded program in the country, plus Big Tech AI teams that never show up on a mobility job board.

Everyone has gone onsite at the same time. Of the 216 roles my board tracks this morning, exactly one is remote. When the entire industry demands relocation simultaneously, the effective talent pool for your req is not "every autonomy engineer." It is "every autonomy engineer willing to move to your metro this year." That is a fraction of a fraction, and it is the single biggest reason your req math feels broken.

The five reasons the req stalls

1. You priced the role against automotive, not against the people actually bidding

Autonomy engineers do not benchmark themselves against OEM salary bands. They benchmark against Big Tech and AI labs, because that is who is in their inbox. If your comp committee approved a band built from automotive survey data, you are opening negotiations 20 to 30 percent under the candidate's mental anchor before anyone says a word. You do not have to win on base. You do have to know the real market price and decide deliberately what you will compete on instead: equity, scope, ship-date credibility.

2. Your loop was designed for software or hardware, not both

Autonomy sits on the seam. A loop borrowed from a software org filters out brilliant systems and controls people who do not grind LeetCode. A loop borrowed from a hardware org bores ML engineers into withdrawing. The fix is one loop, deliberately split: one rigorous technical evaluation in the candidate's actual discipline, one cross-functional session on the seam itself, and nothing redundant. Every additional round past four interviews is a tax you pay in withdrawal rate.

3. Generic sourcing for a domain-specific market

Agencies keyword-match. They send you "computer vision" resumes when you need someone who has shipped perception on an embedded platform with automotive safety constraints. This market is small enough that domain-literate direct outreach beats volume every time. When I sourced directly for AFEELA, knowing the difference between a robotaxi stack and an L2+ consumer program was the difference between a 5 percent and a 40 percent reply conversation. That depth saved the program $1.5M+ per year in agency fees, and the candidates were better.

4. Latency is killing you quietly

The strongest autonomy candidates are in multiple processes the moment they take your first call. Every day between final interview and offer is a day a competitor closes them. Most 90-day reqs are not 90 days of searching. They are 30 days of searching wrapped in 60 days of scheduling gaps, debrief delays, and approval queues. Pre-calibrate the offer before the final round, debrief same day, and get the offer out within 48 hours. Speed is a compensation strategy that costs nothing.

5. Nobody pre-closed the candidate

An offer should be a formality, not a reveal. By the final round you should know the candidate's number, their other timelines, their relocation constraints, and what would make them say no. At AFEELA I made 49 offers and 48 were accepted, a 98 percent acceptance rate against an industry average of 85 to 90 percent. That number was not luck. It was pre-closing: every offer was tested verbally before it was ever written.

What good looks like

Those are my numbers from building the AFEELA U.S. team as a sole recruiter: 48 hires across 9 functions in 12 months, at a 42-day average time to fill against an industry norm of 60 to 75 days. The playbook was not exotic. Price the role against the real bidders. Build one loop that respects both halves of the discipline. Source directly with domain fluency. Kill latency. Pre-close every offer.

Your milestone does not care whose fault the open req is. But the fix is rarely "more resumes." It is repositioning the req against the market that actually exists: smaller than you hoped, more competitive than you budgeted for, and very winnable for the team that moves fastest and knows the domain best.