Mar 2026
Tesla FSD: Human-AI Collaboration in Decision Making
Overview
This exploration is based on my everyday experience with Tesla Full Self-Driving. Rather than evaluating the system itself, it explores how decision-making shifts when control is shared between a person and an AI system. It reflects on how judgment develops through experience rather than assumption, and why people remain an active part of the decisions being made.
Why I Finally Tried It
I have owned a Tesla for more than two years, but I only started using Full Self-Driving about two months ago. Before that, people had recommended it many times, but my answer was always no. It simply did not feel safe. Even though I work in the tech industry, I was surprisingly resistant to trying it.
A close friend who recently bought a Tesla kept encouraging me to give it a try. At some point, I caught myself thinking it was strange: I work in tech, yet I was refusing to try something without really understanding how it worked. That realization made me approach it with a more open mind. After I started using FSD, I also learned that another close friend's uncle had been using it since its early US release without any major problems. Using FSD made me realize how much of my hesitation had been driven by perception rather than experience.
How It Feels to Use It
Using Full Self-Driving does not make me feel clearly like a driver or a passenger. Instead, that role shifts depending on the situation. On unfamiliar roads, I feel more like a passenger. On familiar routes, such as my commute, I still feel like the driver.
During the first week of using FSD, I took over frequently. That phase felt less about trust and more about learning how far the system could actually go. Sometimes it signaled a lane change before I had even considered one myself. Those moments made me realize it was constantly observing its surroundings, even when its decisions did not match my own.
My Role Shifts With Context
Daily Commute
driver
New Route
driver
Light traffic
driver
Busy Traffic
driver
Time Pressure
driver
No Time Pressure
driver
Learning Where to Step In
At the beginning, I took over very often, not because something had gone wrong, but because I was still learning how much control to give the system. Over time, I started to recognize patterns. What mattered was not whether the system made the right decision, but whether I understood what it was likely to do next. As its behaviour became more familiar, I found myself taking over less frequently. It's not because I trusted it completely, but because I could better predict how it would respond.
Before using FSD myself, I used to notice Teslas hesitating in the middle of the road and found it confusing. Now I realize those moments are probably just people like me taking over the control with the system which I do it almost every day.
01
Frequent takeovers
02
Recognize patterns
03
Predict behaviour
04
Selective takeovers
When the System Asks for Feedback
Every time I intervene, the system asks whether I want to submit an anonymous recording of what happened. Because I understand how those reports help investigate specific situations and improve the system over time, submitting feedback has become almost automatic. Sometimes, when I forget to report an issue, I even find myself hoping I encounter the same situation again so I can submit it.
Intervene
Tesla asks
Submit report
Training data
Future system behaviour
A close friend of mine also uses FSD, but she has never submitted a report. She simply has no reason to engage with it in that way, so it never crosses her mind. Seeing that difference made me realize who actually contributes feedback. The system is learning from a very specific group of users, and I happen to be one of them. Who chooses to respond, and who does not, shapes what gets reported and what never does.
The Moments I Take Over
I take over most often during lane changes on busy roads. In heavy traffic, FSD tends to wait until it is fully confident there is enough space, making lane changes harder to complete. I also intervene when it brakes later than I expect or hesitates at yellow lights. These moments are rarely about safety, but they reflect a different threshold for making time-sensitive decisions.
Some takeovers are about safety, while others simply reflect driving style. The moments I find most interesting, however, are not about either. Sometimes I choose to yield to another driver simply because it feels like the right thing to do. The system follows its own rules and does not account for that kind of informal coordination. Those moments are not about efficiency or correctness. They are about reading intent and responding to it.
Driving style is also deeply personal. When to yield, how assertive to be, or what feels comfortable in tight situations varies from person to person. Although FSD offers different driving modes, I rarely notice a meaningful difference between them. Some takeovers are not about safety but simply reflect how I prefer to drive. It is just not driving the way I would.
What Changed My Perspective
01
Managing Uncertainty
Leaving control to AI sometimes felt less like trust and more like choosing not to overreact.
02
Predictability Over Correctness
What mattered was not whether the system made the right decision, but whether I understood what it was likely to do next.
03
Adapting to AI
The more decisions we share with AI, the more our own judgement and behaviour may gradually adapt in return.
Final Reflections
What stands out to me is not just how advanced the technology is, but how much openness it asks from the people using it. I avoided FSD for a long time without fully understanding why. That experience reminded me how easily assumptions can shape our willingness to adopt new technologies. Not everyone will have the opportunity or willingness to experience that shift, and that gap matters. It influences who participates, who gives feedback, and ultimately whose behaviour these systems learn from.
Working alongside FSD also changed how I think about designing AI systems. Rather than replacing human judgement, the most effective systems create space for people and AI to contribute differently. The goal is not to remove people from the decision-making process, but to design better collaboration between human judgement and intelligent systems.
Read Next 👉