A battered Grand Cherokee with wood paneling, a cracked windshield and chunks cut from its side-view mirrors serves as a test vehicle for Stellantis’ hands-free driving system. The openings are where cameras are mounted, Joydip Saha explains from behind the wheel.

Saha, the global engineering manager and owner of Stellantis’ L2++/hands-free door-to-door supervised automated driving system, calls it “the most broken-up car.” A yellow box on the dashboard gives a human safety driver control through an industry-standard “kill switch.” The switch disconnects power to the advanced driver assist system module, or ADAS, preventing it from controlling the vehicle. The driver then has full control.

The STLA AutoDrive L2++ system embedded in the vehicle beeps when it needs Saha to intervene, much like a dog wearing a smart collar. During a test inside the Chrysler Technology Center in suburban Auburn Hills, Mich., the system becomes confused while navigating the complex parking structure, and Saha has to take the wheel.

The campus was built by Lee Iacocca, the former Fortune poster boy who led Chrysler in the 1980s. Construction began on the 500-acre site in 1986, and the facility was dedicated in 1991 at a cost the company puts at about $1 billion. With its headquarters tower added in 1996, the complex covers 5.4 million square feet, making it the second-largest building in the United States by floor space, behind only the Pentagon.

Stellantis still runs its North American engineering operations from Iacocca’s building. Employees continue to describe the arrangement as being “all under one roof.” The interior retains an early-1990s design, with narrow escalators surrounding an octagonal skylight and a diamond-shaped parking garage that is both distinctive and difficult to navigate.

Auburn Hills’ suburban roads prove more suitable for testing. Saha’s development vehicle smoothly navigates an intersection where two Michigan drivers make illegal multi-lane changes within seconds of each other.

Saha has worked on autonomous-driving systems longer than most of his colleagues. He worked on the Chrysler Pacifica that Stellantis supplied for Waymo’s first robotaxi service in Arizona, then managed a group that integrated Level 4 autonomy with partners including Aurora and Lyft. He still test-drives every new software build himself, sometimes twice a week, because he considers any problem experienced by a customer to be his responsibility.

Saha said his team built the Cherokee’s hands-free system in six weeks with the help of AI tools. Similar work once took a year or as long as 18 months. He nevertheless imposes limits on how engineers use AI.

“I will not let people use AI to architect stuff because they will forget how to debug stuff,” Saha said. In a safety-critical industry, he added, engineers must be able to answer every question about the systems they build.

Across two days of interviews at the technology center, Stellantis engineers told Fortune that AI had reduced some yearlong projects to weeks and some overnight simulations to minutes. The company wants to use that speed to shorten development of its STLA Brain computing platform and STLA SmartCockpit software. More broadly, Stellantis is seeking to reduce vehicle-development time from four years to 24 months, beginning with two models in 2027.

Almost every engineer who described faster work also emphasized the need to check the machine’s output. Stellantis’ challenges extend beyond development speed. The company has 14 brands, a European business that is losing money and a second-quarter adjusted operating margin of 1.8%. Chief Executive Officer Antonio Filosa has set a margin target of 7% for 2030.

Wall Street is not waiting. Stellantis is the Detroit Big Three automaker most exposed to Europe, which accounted for about 38% of its 2025 revenue. General Motors sold its European brands, Opel and Vauxhall, in 2017 to PSA, one of the companies that later merged to create Stellantis.

Europe is also moving toward electric vehicles more quickly than the United States. Battery-electric cars represented 17.4% of new EU registrations last year, according to the European Automobile Manufacturers’ Association, or ACEA, compared with 7.8% of U.S. sales, according to Cox Automotive. The European Union has also set 2035 as the point when new cars sold there must be zero-emission at the tailpipe, a rule that applies to every automaker selling in the region.

Stellantis had also made a major bet on electric vehicles in North America. It took the largest hit among Detroit automakers after the $7,500 federal EV tax credit ended on Sept. 30, 2025. In February, Stellantis recorded a write-down of roughly €22.2 billion, or about $26.5 billion. That exceeded Ford’s $19.5 billion charge in December and the $7.6 billion that GM recorded across two charges.

Stellantis shares fell 25% that day, and the effects have continued to weigh on the company. The stock is down nearly 60% year-to-date.

Mel Roberts, Stellantis’ director of North American infotainment, plans vehicle features five to eight years in advance. She said AI was not the biggest surprise of the past several years.

“What I couldn’t anticipate is just the complete collapse of the electrification ambitions and how that really did affect all of our future plans,” Roberts said. “It trickled into everything.”

Detroit’s European automaker

Stellantis is the only member of Detroit’s Big Three that is legally headquartered abroad, in the Netherlands. The company was created through a pandemic-era merger in 2021 between Fiat Chrysler and the PSA Group, bringing European brands including Peugeot and Citroën together with Fiat, Jeep and Ram.

The company’s international character was evident during Fortune’s interviews. Executives born in France, Mexico and Bangladesh were encountered within minutes of one another, along with a Michigan employee returning from several years stationed in Paris.

The mergers also left Stellantis with a vehicle-development system built by committee. Ben Zawodni, a software engineering manager, is working to consolidate hundreds of electronic control units into a central computing system called STLA Brain. Each electronic control unit is a small computer with a specific function. Consolidating these small computers into central platforms reflects a wider industry shift toward software-defined vehicles, in which more of what a car does lives in code rather than in dedicated hardware.

Over the decades, Zawodni said, the vehicle’s architecture evolved to mirror the supply chain. “We were bolting on features,” he said.

Amanda French, who leads Stellantis’ platform teams, is addressing a similar problem at another level. The company historically built separate vehicle platforms, each with a fixed size. Its new STLA One platform is designed to stretch from small B-segment cars to midsize D-segment vehicles.

Stellantis says STLA One is intended to grow into a mega platform supporting more than 30 models and targeting over 2 million units by 2035. Asked whether building it felt like playing with Legos, French replied, “Yeah, with a lot more people and emotions involved.”

‘Not there yet’

Ned Curic, Stellantis’ Chief Engineering & Technology Officer, is a talkative and warm executive who describes himself as “intense.” He joined Stellantis from Amazon, after earlier stints at Microsoft and Toyota.

Curic said Stellantis is not yet where it needs to be. Some teams have achieved “phenomenal successes,” he said, but “as a whole organization, we are not there.”

He is frustrated by the complexity the auto industry has accumulated. Modern vehicles can have “five screens, like digital noise, like 350 features,” Curic said. Automakers have historically tended to “checklist the heck out of the features” without determining how customers actually use them.

Curic said he learned at Toyota that “the highest quality part that you have in a car is the part that you don’t have in a car.” Stellantis’ driver research, he said, has found that most people prefer a mechanical stick to a button on a screen.

At Amazon, Curic received emails from Jeff Bezos forwarding customer complaints. The body of each message consisted of a single question mark.

“I never found it stressful,” Curic said. “Never.”

One executive would take responsibility for the problem, he explained, while others would each write a few paragraphs to help resolve it.

Curic also brought ideas from Amazon and Toyota to Stellantis, including small teams based on Amazon’s “two-pizza rule” and a single owner for each system. The goal is to prevent engineers in separate groups from blaming one another when something fails.

“If you see something that’s wrong, you just pick it up,” Curic said. “I own it.”

The most difficult idea he is promoting is that “the best thing that you can do for your career is to make yourself redundant.” Curic said he has done so several times by putting a team in good shape, moving on and starting something new.

For that approach to work, he said, Stellantis must ensure that new opportunities are available to employees. Leaders also need maturity and confidence. Asked whether Detroit’s long period of difficult conditions has made workers defensive about change, Curic said there is always some fear, but change is increasingly accepted “and actually sought after.”

The executive urging engineers to make themselves redundant is also hiring. At the Paris auto show in October, Curic told Bloomberg that Stellantis would add “multiple hundreds” of engineers in Europe and the United States during the year.

The starting gun

The pressure falls on engineers such as Zawodni and French. Both are working on Stellantis’ margin problems by simplifying vehicle wiring and bringing manufacturing processes into closer alignment, with the aim of cutting costs and improving profits.

“The AIs aren’t going to tell us what the new platforms are,” Zawodni said.

He began working on consolidating central computing in Stellantis vehicles during the COVID-19 pandemic. Engineers moved fully remote, and the company established laboratories in India, Poland and North America that could be accessed from anywhere.

A proof of concept for a global observability platform, which monitors software deep inside a vehicle’s computers, took one week. Before agentic coding tools, Zawodni said, comparable work would have taken about six months. Defects that once took two to three weeks to fix were reduced to two or three days.

“We knew that we had to really reimagine all of our time,” Zawodni said. Asked whether that was frightening, he replied, “How exciting, right?”

French, an engineer from southern Indiana, was enthusiastic about AI but offered a more measured assessment. She said human judgment “cannot be replaced.” AI helps her assemble requirements and bills of materials from regional systems that were never designed to communicate with one another because Stellantis is “made up from a number of different companies.”

She compared AI with an interactive version of Cliffs Notes, something she wished she had in school.

Zawodni said his engineers had responded unevenly to the advances. “I think there was a starting gun, and that starting gun went off,” he said. “Some people heard it, and some people didn’t.”

He described the reaction among his own engineers as mixed. “Engineers are really afraid of losing their core skills,” he said. Some want to understand why a semicolon in a programming language behaves as it does. Zawodni called that “a valid curiosity,” but said it would not by itself help Stellantis reach its goals. He wants engineers to become architects and eventually chief architects.

The tools have not reduced Zawodni’s own workload. “Agentic tools, they’ve made our work so much easier,” he said. “So why am I working so much harder?”

He estimates that he works at least twice as hard as before. “I feel a tremendous sense of urgency,” he said. “I really hope that we can transition responsibly … especially our people.”

Choosing battles

Anne Laliron, a French engineer who leads Stellantis’ roughly 400-person technology research group, gave up her office after COVID so employees could always find her. Her role includes deciding which major technological bets the company should make.

Stellantis cannot afford to pursue everything, Laliron said. It designs some battery cells internally and works with startups on others.

“If you want to embrace everything,” she said, “you are at risk of spreading your effort.”

Partnerships allow Stellantis to pursue technologies beyond its internal capabilities. The company works with British startup Wayve on perception software for hands-free driving, with French startup Tiamat on sodium-ion batteries and with Zeta Energy on lithium-sulfur batteries. Both chemistries are alternatives to the lithium-ion cells that power most electric vehicles on the road today. It also owns a stake in Chinese electric-vehicle maker Leapmotor and sells Leapmotor vehicles outside China through the joint venture Leapmotor International.

Laliron said partnering is deeply established at Stellantis and less common among other automakers. Curic wants a culture in which “it doesn’t have to be invented here.” He said the relationship with Leapmotor provides benefits in both directions: “We learned a lot… we also taught them a lot.”

Laliron said Stellantis follows a similar approach with batteries: design them internally, partner with startups and leave mass production to gigafactory partners.

One of the company’s largest bets sits on a nearby stage: a battery-electric Dodge Charger prototype equipped with a solid-state battery from Factorial, a Massachusetts startup that has worked with Stellantis since 2021. Solid-state designs replace the liquid electrolyte inside conventional lithium-ion cells with a solid material, a technology much of the auto industry is pursuing.

The two companies operate as one team. Factorial co-founder Siyu Huang said the startup developed the technology with $250 million in capital, a fraction of what some competitors have spent. The vehicle worked the first time the companies turned the key.

Huang nevertheless estimates that the technology is three to five years away from automotive production. Laliron cuts about 10% of her group’s projects each year when they fail to produce results.

“It’s always painful,” she said.

Still one car a day

AI-driven speed gains are real, but they remain small relative to Stellantis’ broader challenges. Arturo Guzman was among the first employees to introduce AI into the company’s aerodynamics work, about five years ago, after a startup struggled to explain the technology to him during a side conversation at a conference.

He now explains it to vice presidents by removing the mystery. “There’s no miracles. There’s no magic,” Guzman said. “For us being engineers, all this is a Python code.”

Using his team’s model, a design change to the front splitter of a Ram pickup produces a drag estimate in about three minutes. A full simulation previously took 14 to 24 hours.

Guzman compares the system with an intern. To younger engineers, he uses a different comparison: R2-D2 from Star Wars. “You just give it a task and come back and see what it did,” he said.

The time his team saves is devoted to other work, Guzman said. The group creates new guidelines, handles requests from across Stellantis and searches simulation data for potential patents.

“Now I tell them, just send me the data,” he said.

The wind tunnel itself has not accelerated. The equipment still generally tests “one car a day,” said Steve Kubiak, a Michigan native who began working at Chrysler in 1992 and has supervised the tunnel for about 10 years.

Asked whether the process had changed, Kubiak replied, “No, it’s always been the same.”

The analysis is faster even when the physical testing is not. Kubiak’s colleague Todd Lounsberry gives AI data from the previous 20 test runs and asks what patterns it finds.

Asked how someone becomes a wind-tunnel manager, Kubiak answered, “First you get a wind tunnel.”

Other engineers are keeping AI within strict limits. Mohammed Malik, who runs Stellantis’ driving simulators at the Automotive Research & Development Centre in Windsor, Ontario, said his group is “keeping it in the box.” Even so, the technology has changed his work.

Malik’s simulator resembles a full-size, immersive version of Mario Kart, the Nintendo racing game in which characters choose vehicles and tracks and avoid obstacles to reach the finish line. His work has been transformed more by the simulator than by AI.

Engineers once tuned a vehicle on a computer, built a prototype and only then learned how it felt. Now a driver can experience a computer model before a physical vehicle exists, while Malik can change springs during a test. He can test 10 tires in an hour, something that cannot be done with a prototype.

His team has scanned about 20 roads, from racetracks to local streets, in 10-centimeter squares. It exchanges road models with colleagues in Brazil and Italy instead of shipping vehicles across oceans.

Malik showed a video of a full-size Stellantis vehicle body without wheels. It was suspended on what appeared to be mechanical spider legs, moving up and down and back and forth in front of a giant television screen displaying a track.

Asked whether he had played Mario Kart, Malik said a student had built him a track of his own. “You can drive the Rainbow Road if you want,” he said, beaming from behind his bushy mustache.

AI currently has a narrower role in the simulator. A single run produces thousands of data channels. When a vehicle hits a bump and “swirls a little bit,” Malik asks AI which channel should be examined.

His larger ambition is to feed an agent every test driver’s scorecards from every region, ask it “what does it take to make a Jeep,” and allow it to operate the simulator until the vehicle feels right.

For now, he said, Stellantis is moving slowly and wants to retain some form of control. “Some of the stuff we want to do, we can do today,” Malik said. “And some of the stuff we want to do, we cannot.”

The larger surprise has been the differing willingness of employees to adapt. Veteran test drivers who spent 10 or 20 years working with physical prototypes were slow to adopt the simulator. Younger engineers with racing setups in their basements embraced it quickly.

Laliron has observed a similar pattern. Experienced engineers carry memories of earlier failures, she said, and sometimes “their brain is closing doors.”

Some employees have accepted work that is better without necessarily being faster. Roberts said her AI research tool returns 10 to 15 sources, all of which she must check.

“I don’t think it saves time,” Roberts said, although it makes her work more “robust.”

One more cycle

Alison Rahm, Stellantis’ chief vehicle engineer for midsize and large SUVs, has seen technological change before. She is returning to Michigan after two years managing the company’s European car lineup from Paris. She recalls a 1997 staff meeting when her boss announced that employees would receive something new called email.

“All of us were in a total panic,” Rahm said. They returned to their desks and tried it anyway.

AI now performs checks that engineers once completed by hand, Rahm said, including reviewing a list of about 1,000 engineering design practices. That gives engineers more time to ask whether the rules themselves remain correct.

Rahm is not ready to stop double-checking the technology. For years after virtual crash testing was introduced, engineers continued crashing real cars into walls to confirm that the virtual scenarios were accurate.

Asked how long it would take to trust AI in the same way, Rahm paused.

“I think we need another life cycle to prove to ourselves that our assumptions were right,” she said. After noting that different products have different life cycles, she considered the question further. “I would say four years.”

That would put the milestone in 2030, the year Filosa has promised a 7% margin.

Saha drives back into Iacocca’s garage and again takes control of the Cherokee as the diamond-shaped structure comes into view. The automated system still needs a guiding hand to return home.

As the vehicle completes a smooth parking maneuver, Saha discusses when the AI transition will become clearer. He says he is a Game of Thrones fan and that his favorite character is Robb Stark. Although his work is based on logic, he likes fantasy because it encourages him to be “a little bit philosophical” with the other side of his brain.

“Creativity is very important,” Saha said.

Humans have adapted and thrived by finding creative solutions, he said. “We were never the strongest animal in the world,” he said, echoing the style of Game of Thrones author George R.R. Martin.

Saha suggested that several hundred years of industrialization had sometimes trained people to think primarily in terms of processes. “My understanding is we’re going to leave behind the process-oriented world because AI is going to help us. And my personal belief is that’s actually good.”

“There will be some loss,” Saha said, as there is with any technological shift. But humans have repeatedly adapted to the natural world through creativity.

“People are going to figure out how to tame AI, too,” he said.

This report was updated to clarify Ned Curic’s and Joydip Saha’s job titles: Chief Engineering & Technology Officer and owner of L2++/hands-free door-to-door supervised automated driving, respectively.

Originally published on Fortune.com: https://fortune.com/2026/11/01/stellantis-detroit-ai-best-thing-make-yourself-redundant-chrysler/