Self-driving used to be a movie set piece. Now robotaxis run paid routes in San Francisco, mining trucks haul ore from GPS-mapped pits without drivers, and agricultural equipment works fields with no one in the cab.
For all that momentum, the engineering and operational realities behind putting a self-driving stack on public roads stay largely invisible.
Akshat Gattani has spent most of his career inside that hidden layer. First at a major robotaxi operator, and now as a technical program management leader at a Sunnyvale-based company powering physical AI across automotive, defense, trucking, mining, construction, and agriculture.
He's one of a small number of engineers who has worked the autonomy problem from both sides: building and scaling a driverless fleet at one of the field's most ambitious operators, and then building the shared infrastructure the rest of the industry now runs on. That dual vantage has given him a clear view of what the autonomy industry actually needs to deliver on its promise.
From Diesel to Driverless
Gattani's training is automotive engineering, top to bottom: a bachelor's degree, then a master's from a top rated university. His early professional work centered on reducing diesel emissions and navigating global regulatory requirements for commercial vehicle products.
The move to California was a deliberate escalation. Where most automotive engineers follow a conventional path through established OEM programs, Gattani relocated to join a company attempting something no legacy automaker had pulled off at scale: fully driverless commercial operations on public roads, with no safety driver behind the wheel.
His initial work focused on the integration layer. Getting autonomous software to function reliably on physical hardware is the step that determines whether any autonomy stack can translate from a simulated environment to an actual car. From there, he moved into fleet scaling, coordinating on vehicle manufacturing and overseeing the city-by-city expansion of driverless operations.
The fleet grew substantially during his time there, spanning operations across San Francisco, Los Angeles, Phoenix, Houston, and Dallas. Each new city introduced a distinct set of engineering and operational variables. Road geometry, traffic patterns, and local signaling conventions all differed in ways that forced the stack and operations teams to rebuild their original plans.
That's when he realized the scaling work for this technology is as much a logistics problem as a software one.
The Safety First Mindset
The more instructive part of Gattani's time at this robotaxi company, by his own accounting, isn't the fleet he helped build. It's the safety-first mindset he developed over three years of hands-on testing and deployment work.
The testing sequence is deliberately graduated. Closed-course testing with a safety driver comes first, setting a baseline performance threshold. Then comes limited public-road testing with a safety driver present, followed by constrained driverless operations in low-risk conditions, typically the overnight hours between midnight and 4 a.m. in low-traffic corridors. Only after all of those gates are cleared does expansion toward full 24/7 operations become a possibility.
Gattani's team worked closely with the closed-course testing group during the early development of a next-generation vehicle platform. He was tasked with finding integration issues early in development and working with engineers to fix them before any vehicle moved to public-road testing. Simulation tools played a central role in that triage process, allowing the team to identify and work through issues before they reached physical testing.
"Early in closed-course testing, we'd run into issues like phantom braking that pushed back platform readiness," Gattani says. "Every phase has quality gates. If we didn't hit them, we delayed. The platform had to be safe before it went on public roads."
He has seen firsthand what happens when that discipline breaks down, as a single public failure, regardless of how much development preceded it, can end an entire program. "The stakes of not delivering a safe product are extremely high," Gattani says. "In my space, moving fast and breaking things usually means the company breaks too."
That's a sharp counter to the move-fast ethos that defines so much of Silicon Valley, applied to a domain where a single systemic failure can be catastrophic and irreversible.
The Picks and Shovels Bet
Three years into working in this field, Gattani made a bet about where the next decade of autonomy progress would actually come from. Not from any single operator's fleet, however large, but from the layer underneath all of them: the data, simulation, and development infrastructure every serious autonomy program now relies on. So he left.
"Instead of being in a vertically integrated company that goes all the way from development of the stack to deploying robotaxis out on the road," he says, "I wanted to work in the space where you're developing what we would call the picks and shovels of the industry."
He now works on the tooling side of the industry, focused on the data, simulation, and development infrastructure that serious autonomy programs depend on. From global OEMs to specialized L4 autonomy stack developers across automotive, trucking, and defense, customers across those sectors use the platform to develop and validate their own autonomy programs.
Gattani's current role as a technical program management leader is to make complex, cross-functional programs actually ship. In practice, that means translating market needs into roadmap priorities, defining what "done" looks like before teams start executing, and surfacing the integration dependencies that quietly derail large programs. When Team A's deliverable hinges on a prerequisite Team B hasn't built yet, his job is to flag it weeks ahead, not after the slip.
He also owns the metrics layer for the entire organization, defining how progress is measured across teams and whether shipped features are actually changing customer outcomes.
The Case He Keeps in Mind
The argument Gattani returns to most often, when describing what drives the work, is empirical. Published safety data from autonomous vehicle deployments has shown that self-driving systems can outperform human drivers on key metrics, removing accident causes like drunk driving, phone distraction, and fatigue that account for a substantial share of traffic injuries and deaths each year.
That safety case, in his view, is the outcome the technology can deliver at the largest scale: a measurable reduction in the public health burden human driving error currently imposes.
He also sees widespread deployment cutting urban car ownership, which would ease congestion on city streets and return commuting hours to more productive or restorative use. In his framing, these are the logical downstream effects of a system proven to work and steadily expanding its operational range.
That thinking extends well past passenger vehicles at his current company. Trains, drones, warehouse forklifts, agricultural tractors, and mining trucks all fall within the scope of what the platform is built to support.
The same principle shows up in how Gattani runs his teams. He's built his organization on a clear premise: the hardest part of leadership is hiring well, and once you have, the rest is staying out of the way.
"You have to trust the people, and trust yourself that you've hired the right ones," he says. "If you've done that part well, the job is to let them do the work."
The measure that matters, by Gattani's reckoning, is the gap between how many people human drivers kill each year and how many people machines will, once the discipline he spent a decade learning is built into every stack on the road.
Charting The Road From Simulation to Streets
Akshat Gattani's career traces a line through nearly every layer of the autonomy industry, from the foundational engineering work of integrating software with hardware, to the operational reality of scaling driverless fleets in American cities, to the safety discipline that determines when a vehicle earns the right to be on a public road, and now to the simulation infrastructure that gives this industry the tools to do that work more rigorously. The thread running through all of it is a belief that the technology's most meaningful contribution isn't a faster commute or a more convenient ride, but a measurable reduction in the harm that machines cause when human error is removed from the equation.
