A Year and a Half with Tesla FSD in the Bay Area

For roughly the past year and a half I have used Tesla’s Full Self-Driving (FSD) system almost every day on the exact same morning route to my daughter’s school. That consistency makes it easy to see how FSD behaves under different profiles and software releases.

Driving profiles I rely on

Tesla labels its driving styles differently across versions, but they map to levels of aggressiveness. I live almost entirely in Hurry (≈80% of my driving) with a dash of Mad Max (≈20%). I rarely touch Standard and have essentially never used Chill.

Hurry feels like a careful new driver. Lane changes only happen when there is a generous opening, which keeps things polite but can be frustrating when traffic is dense.
Mad Max acts much closer to my own style when I am in a rush. It takes smaller gaps, changes lanes decisively, and keeps momentum without feeling reckless. The first time I toggled it back on after running late, the car immediately began threading through traffic to preserve speed.

Speed behavior differences

Each profile also enforces its own relationship with the posted speed limit, on top of whatever offset you allow in settings:

Standard hugs the exact limit.
Hurry nudges a little above it.
Mad Max is comfortable going noticeably faster if you configure a higher offset.

Because I drive identical roads at identical times, those differences jump out quickly.

Everyday Bay Area commuting

On Bay Area roads I am on Autopilot or FSD about 99% of the time. The experience keeps getting smoother, and Tesla continues to add useful conveniences: automatic HOV-lane decisions when the occupancy is detected, optional toll-lane handling, and generally more natural lane changes and merges than earlier builds.

Limitations and edge cases

Despite the progress, some weaknesses feel persistent:

Glare: Direct sunlight into the cameras still triggers immediate takeover requests.
Random lane picks: Maybe once or twice a year the car drifts toward an unexpected lane without clear reason.
Long-haul weirdness: A friend driving across Utah had FSD suddenly aim for a neighboring truck after hours on the highway, mirroring scattered reports online. My hunch is that extremely long drives eventually encounter a low-probability bug.

I avoid most of those issues because Bay Area conditions match a huge amount of Tesla’s training data, but the imperfections remain.

Why vigilance still matters

Complex intersections can confuse the planner, occasionally leading to late lane changes or route corrections. None of my incidents have felt truly dangerous, yet they reinforce the obvious: FSD is still advanced driver assistance, not a license to tune out. The jokes about napping behind the wheel are still just jokes.

Robotaxis vs. Tesla’s approach

Companies like Waymo operate fully driverless taxis inside tight geofences. Tesla lets you run FSD almost anywhere but still insists on supervision. It is simply a different risk posture—Tesla accepts a broader domain by keeping the human in the loop, while robotaxi operators restrict geography so they can remove the driver entirely. Accident data so far suggests both approaches can stay quite safe within their boundaries.

Long-distance trips still impress

One of my favorite FSD moments was a five-hour drive to Yosemite. The system handled highways, rural stretches, and winding park roads—even without cellular coverage—because it relies on onboard perception. Compared with my manual driving, FSD kept speed far steadier, especially through curves. Ironically I felt more nervous when I briefly took over.

Looking ahead

Every release tightens up behavior a little more. Interventions are less frequent, lane changes are smoother, and the car simply feels more confident. We are not at full autonomy yet, but the trajectory is obvious: FSD keeps shrinking the set of situations where I need to jump in. Until it reaches true independence, though, it remains the best co-pilot I have ever had—not a replacement for me.