Silicon Valley is full of comfortable offices with gleaming meeting rooms and on-site baristas. So when early HappyRobot engineer Ari Polakof decided to strike out and start his own company, he didn’t exactly expect to wind up tapping on his laptop in a service bay alongside mechanics.

“Very noisy, impossible to concentrate,” he laughed.

But that’s part of the story of how Polakof’s new AI-for-car-dealerships startup Flai got off the ground last year.

Founded by Polakof and his brother Alen (also from HappyRobot) at Y Combinator, along with former Netflix data scientist Juan Alzugaray, Flai is one of a number of startups trying to use artificial intelligence to improve the experience of buying, selling, or servicing a car at a dealership.

The co-founders have built software from the ground up that’s specially tailored for the car dealership environment. And it’s “omni-channel,” as Polakof said in an interview, meaning it can handle phone calls (using voice agents) as well as emails and texts (using large language models).

Now, Flai has closed a $4.5 million seed round to try and scale up what they’ve built. The round was led by Liz Wessel at First Round Capital, and included funding from YC, RedBlue Capital, Joe Montana’s Liquid 2 Ventures, and Innovation Endeavors.

Flai is not the only one doing this kind of work. Earlier this year, another YC-backed startup called Toma announced it had raised $17 million from the likes of a16z and Yossi Levi, the car industry influencer known as Car Dealership Guy. A swarm of others are trying to build similar products, while legacy interactive voice response companies — the ones who make the phone tree software we’ve all become overly familiar with — look to keep pace.

Polakof, who is CEO, said Flai’s offering is different in that it’s not using off-the-shelf voice tech. The startup has built essentially everything from scratch, which in particular has made its voice agents impressive enough that Polakof claims it’s already convinced a few dealers to switch from other companies.

Besides, he said, there is a lot of room for competition in the market. There are thousands of car dealerships and service centers across the United States, and the majority of them have the same problem: they can lose potential customers if their phone lines are tied up.

After getting off the ground last year, the Flai team realized the best way to train their AI, and to start racking up customers, was to start showing up at those dealerships.

Sometimes they tried cold calling, or emailing, or LinkedIn messaging, or even showing up at conferences to network. But the bulk of the early work Flai got done was from going directly to the brick-and-mortar locations — around 400 in those early days, by Polakof’s count.

Once Flai struck up these initial relationships, Polakof said they essentially embedded with each dealer. That’s how he found himself in a service bay last year — though, he said, more often it just meant setting up in an empty office.

Polakof said the Flai team spent “every day, all day on the road” during that initial period. “It’s almost very tiring, painful, but I think it’s the only way, and I can’t imagine a bigger company doing it,” he said.

As the startup transitions out of that learning period and into one more focused on growing and making customers happy, Polakof said the small team is still pushing long hours — like so many of their peers in Silicon Valley.

The seed funding will help Flai grow, but Polakof said he doesn’t expect the team to explode.

“We want to work as smart as we can,” he said. “We don’t want to have 100, 200 people working on this right now.” Instead, he said the three-man team keeps each other “accountable” by focusing on tasks that are “100% required, otherwise it’s not worth it.”

In that sense, Polakof isn’t so far from those earliest days working in a service bay or dealership back office — just with less noise and fewer mechanics.