The entire automotive industry is racing to build autonomous driving technology, but few teams have been working on the problem longer than the company now known as Alphabet. Seven and a half years into the project, Alphabet’s X division said today that its cars have now driven 2 million miles on public roads in fully autonomous mode. That’s a collective 300 years of driving, an X spokeswoman said, during which the cars have navigated hundreds of millions of interactions with vehicles, pedestrians, and cyclists.

To mark the occasion, Alphabet invited me to take a ride in one of its test cars with Dimitri Dolgov, who leads software development for X. Dolgov, who joined Google in 2009 to work on the project, said the company is having gradual success in making the ride feel smoother and safer. It’s identifying edge cases and refining how the cars handle them. "Every day in simulation we re-drive about 3 million miles," he said. "Every time we make a change to the system, we can pipe it through all the data we’ve collected, and evaluate it against that data set. That’s kind of a big deal for us."

It’s still not close to becoming a commercial product — that will only come "when it’s ready," Dolgov said — but the technology is becoming increasingly sophisticated. We hopped inside one of the company’s modified Lexus RX450 vehicles and set out on the streets of Mountain View. Dolgov encouraged me to look for the subtleties in our ride: the way the car successfully navigated a 4-way stop when there was a vehicle waiting at each stop sign, for example, or how it slowed when a car in front of us unexpectedly began to reverse.

It was the third time I had been a passenger in one of Google’s autonomous vehicles. Once a year for the past three years, the company has invited me to its headquarters in Mountain View to get an in-person report on its progress.

It’s a time of transition for Alphabet’s self-driving car efforts. The project is close to becoming a standalone business, the company said in April. At the same time, it recently lost some of its top talent, including Chris Urmson, its longtime technical lead. It’s still unclear how Google will brings its cars to market — who will manufacture them, where they will be available, and how you will book a ride in them.

Meanwhile, other companies have accelerated their efforts to build autonomous vehicles. Journalists taking rides in self-driving cars has become a public-relations genre unto itself; in the past two years The Verge has ridden in fully or partially autonomous vehicles from Mercedes, Ford, Uber, Comma.ai, and a big rig known as the Freightliner Inspiration. Meanwhile, Volvo has announced a public test of self-driving cars with 100 average citizens next year.

Google first tested its cars primarily on highways, where they interacted mostly with other cars. In 2014 the company rolled them out on city streets, where they had to fight for space with cyclists, pedestrians, and freight trains, among other obstacles.

That spring I took in a ride modified Lexus RX450h, and I was struck by how ordinary the ride felt. It made minor adjustments for passing cyclists and pedestrians, and slowed to scan for red light runners whenever the light turned green. It moved a bit more gingerly through the world than the human drivers I know, but if I closed my eyes I might not have noticed a difference.

The next year, Google invited me to an obstacle course. It had set up a series of dangers on the roof of a converted shopping mall, and put me inside one of its Smart Car-like prototype vehicles. During a five-minute demonstration, our tiny mobile gumdrop of a vehicle successfully dodged a pedestrian, a sedan that cut us off, and a cyclist who veered in front of us. (The prototype successfully interpreted the cyclist’s hand gesture signaling a left turn.)

This year’s ride was much more tame by comparison. We hit the streets of Mountain View at rush hour, and spent about 20 minutes navigating a winding route around the X campus. The speed limits never rose about 35 mph, and neither did our Lexus. X’s autonomous vehicles are Boy Scouts, meticulously following every posted sign and instruction. When lights turn green they wait about 1.5 seconds before proceeding into the intersection — a hedge against red-light runners. They seem to slow for stop signs roughly 300 feet before a human would.

X had made much of the enhanced smoothness of its rides these days, but I found that the car still moved timidly down the road. It hesitated at odd moments on near-empty streets, the way my grandmother used to. If the car were being driven by a human, you might say it lacked confidence.

Of course, what humans interpret as confidence is actually arrogance. We are, on balance, atrocious drivers compared to machines. An autonomous vehicle moving cautiously through city streets may lack the zip of an experienced driver commuting home — but it’s likely much safer, too. (And, as Dolgov pointed out to me, the software is actually very confident, thanks to its 360-degree laser array and advanced predictive capabilities.)

When these vehicles become commercially available, it may take us some time to adjust to their plodding pace and meticulous rule-following. But I suspect we’ll notice less than we might expect, because we’ll have our faces buried in our phones, or our tablets, or our laptops.

My ride in Mountain View this week ended without Dolgov having to take the wheel even once. That’s what true confidence looks like for an autonomous vehicle. When Dolgov says the software still isn’t ready for a commercial release, I believe him. But it sure feels close to me.