Who Runs Day-To-Day AV Operations? The Autonomous Vehicle Operations Specialist

What Does an AV Operations Specialist Do at 5:40 In the Morning?

At 5:40 the depot has eleven vehicles due out at six and nine of them are ready. One will not clear its sensor calibration check, one has a cleaning vendor who did not show, and a shuttle from last night's shift has an open note that says "hesitated at the construction zone on Fourth." Whoever sorts that out in twenty minutes, and writes the Fourth Street note in a form an engineer can search, is the person this role names.

The traits that matter are unglamorous and they are all about restraint under time pressure. First, calibrated intervention: knowing that a vehicle creeping oddly through an occluded intersection is often the software behaving correctly and looking strange, while a vehicle drifting toward a lane line without an explanation is not. Second, narration under load, meaning the ability to say what is happening while it happens rather than reconstructing it later from memory. Third, a reporting habit that survives a fourteen-hour day.

The tells separating a real candidate from a performed one show up in the specifics of an intervention story. Ask for the last time they took control of a vehicle, any vehicle. A strong answer has a location, a road user, a reason, and a second-guess: what the vehicle might have been about to do, and whether they would make the same call now. A performed answer is a genre scene about quick reflexes and a near miss. Reflexes are not the constraint here. Vehicles brake faster than people do.

The second tell is how they talk about the vehicle being wrong. Someone who has done this work has a stock of small, boring, specific failures: a stale map tile, a delivery truck parked in a way the routing did not expect, a puddle that read as an obstacle. Someone performing the role speaks in categories, about edge cases and disengagements, without a single street name in the whole answer.

The third tell is what they do when nothing happens. Most of a shift is uneventful, and the difference between a good specialist and an expensive passenger is what gets logged during the quiet hours: vehicle condition, cleanliness, charge state, a rider comment worth passing on, a vendor who is slipping. This is the same discipline an autonomous fleet monitoring and response engineer applies from a remote console, done in the vehicle instead.

Which Backgrounds Actually Produce This Person?

Four backgrounds produce good AV operations specialists reliably: commercial drivers with a clean record and a habit of documentation, transit and paratransit operators who have handled passengers in unscripted situations, aviation and maritime crew trained in checklists and handover notes, and technicians from vehicle maintenance who already read diagnostic output without flinching. Each brings a different half of the job.

The commercial driver, particularly CDL-A, is the most direct feed and the postings say so plainly. Outrider hires an autonomous vehicle operator specifically at CDL-A for yard truck work ¹. What that person usually needs to build is the supervisory habit: a driver's instinct is to correct, and this job frequently requires holding still and watching a machine do something you would have done differently. That is genuinely uncomfortable and worth naming in the interview rather than discovering in month two.

Transit operators arrive with the passenger half already solved. A May Mobility shuttle carries riders who ask questions, need a ramp, or become anxious when the wheel turns itself. Somebody who has run a fixed route with a wheelchair lift and a schedule has handled harder human situations than most engineering candidates ever will.

The unexpected backgrounds are worth protecting from your own resume screen. Aviation ground crew and flight operations people are trained in exactly the two habits this job runs on, checklists and handover discipline, and the parallel to an autonomous aircraft flight operator is close enough that candidates move between the two. Military vehicle and UAV crews bring the same. Ski patrol, wildland fire and EMS produce people who make consequential calls on partial information and then write a report about it, which is the whole job described in one sentence. And driving instructors are quietly ideal: their entire trade is supervising a driver they do not control, deciding when to grab the wheel, and explaining afterward why.

What transfers less well than people expect is pure warehouse robotics coordination. Supervising an AMR in a mapped building is a different problem from supervising a vehicle in public traffic with pedestrians, weather, police direction and liability. The postings treat these as separate ladders, and so should you.

Screen for the Operator Who Interrogates the Vehicle's Own Explanation

The specialists who got good at this did it by treating the tools as a claim to be checked. The vehicle's interface offers a running account of what it believes: objects it sees, the path it intends, why it slowed. Fleet software adds summaries and suggested causes, increasingly generated by a model. A specialist who accepts those summaries writes reports that agree with the machine and teach engineering nothing.

Ask how they use whatever assistant or summarizer their last employer gave them, and listen for a checking step. The useful answer sounds concrete: the tool proposed that a stop was caused by a pedestrian, the video showed a plastic bag, and the report says bag with a timestamp. The unhelpful answer describes the tool as accurate and the reports as fast. Speed is not the product here. A wrong cause code, repeated across a month of shifts, sends an engineering team chasing a behavior that does not exist.

The same question applies to the writing itself. Plenty of operators now draft incident notes with an assistant, and that is fine when the person supplies the facts and edits the invented ones back out. What you want to hear is which parts they refuse to delegate: the timestamp, the road user, the reason for the intervention, and their own uncertainty. Anyone who says the assistant handles all of it has not read one of its drafts against the video.

A working screen fits in an hour and does not require a vehicle. Play four minutes of real fleet footage with the vehicle's own status overlay, including one clip where the overlay is misleading, and ask three things: what would you have done, what does the vehicle appear to believe, and write the note. Then read the note for the details that survived. Whether they marked their uncertainty tells you more than whether they matched your answer key.

One thing not to screen for: whether an application was written with AI help. That cannot be determined reliably, and it has no bearing on whether a person can hold a lane decision and a logbook in their head at the same time.

Where Do You Find Them, and How Do You Close Them?

Find them where the deployments already are. The visible employers post openly and by city: May Mobility carries autonomous vehicle operations specialist, specialist II, senior specialist and site operations manager II; Nuro posts tiered autonomous vehicle operator shifts across four metros with a shift supervisor above them; Outrider posts a CDL-A operator role; Waymo carries market operations and fleet vendor performance roles alongside the driving ladder ¹. Their own boards are the honest first stop, including for competitive hiring.

Beyond the boards, the feeders are local rather than national. CDL schools and community college transportation programs in your deployment metro. Transit authority operator pools, especially where a route has been cut. Fleet maintenance shops. Rideshare drivers with several years and a clean record, who are already familiar with the streets you operate on and with being watched by a camera all shift. Municipal permit hearings and state AV advisory meetings put you in a room with the people already working the local details.

Closing this candidate turns on three things, and pay is rarely the first. The first is the ladder: show it. The reason the tiered postings matter is that a driver considering this move is trading a known trade for a young one, and a visible path from specialist to specialist II to senior to site manager is the answer to "what is this in three years." Name the rungs and what moves a person between them.

The second is schedule honesty. Say what the shift pattern really is, whether nights and weekends rotate, and how much notice a schedule change gets. Fleet operations run when the service runs, and candidates from transit already know to ask.

The third is telling the truth about the category. It is young, the ladder is a couple of years old, and a deployment can pause. Say that plainly, then say what the person is being trained on that outlives any one employer: intervention judgment, incident reporting, vendor and depot management, and the fleet tooling itself. The candidates worth hiring have already done that arithmetic. The ones you lose are usually the ones who sensed the uncertainty was being managed rather than stated. That is the same closing problem a construction AI innovation program manager presents in a different industry.

What Should the Role Pay, and Does It Have To Be On Site?

There is no published wage series for this title, and any confident point estimate should be treated as invented. Build the band from the two established roles the posting competes with in your metro: a commercial driver in the same license class for the entry rung, and a first-line transportation or fleet supervisor for the senior and site manager rungs. That is what the tiered ladder is expressing.

Two adjustments follow from the work itself. The reporting and intervention responsibility sits above a straight driving job, so the entry rung typically has to clear the local driving band rather than match it, or you lose candidates to freight in their first quarter. And the senior rungs are supervisory in substance, carrying vendor performance, depot uptime and other people's shifts, so pricing them off a driver band is the most common way one of these searches stalls.

On the wider market: PwC's 2026 AI Jobs Barometer reports an average wage premium of 62% for AI skills across roughly one billion job ads ². That is a market-wide average and not a promise about this title, but it is the pressure behind the tiering. Roles that pair an established trade with the supervision of an AI system are being priced upward, and a fleet job posted at last year's driver band will feel it.

Remote is mostly not on offer here, and it is worth being precise about why rather than defaulting to a policy line. The specialist and operator rungs are in or beside the vehicle, so they are on site by definition, and shift-based. The supervisory rungs are depot-based: the vendor who cleans and charges the fleet, the technician who takes a vehicle out of service, and the site itself are all physical. What is genuinely remote is the console side of fleet response, which is a different role on a different ladder and often hires from these ranks after a couple of years. If a candidate is asking about remote work early, they may be pointed at that path instead, and saying so is a better answer than a policy.

One structural note before you write the posting. Because the category is still forming, your job description is doing double duty as a definition. Write the intervention standard, the reporting expectation and the rung boundaries into the text, because the candidate has no industry-wide picture to fill the gaps with, and neither, honestly, does the industry.

See a sample report >