Elon Musk Says Tesla FSD Is Getting a Memory for the Way You Drive

Tesla’s Full Self-Driving (Supervised) software could learn something more useful than a generic preference setting: the moments when an owner says, in effect, “No, not that way.”

Elon Musk says the car will begin remembering a driver’s specific interventions and matching that person’s individual preferences.

A software version and release date remain unknown. If it works as described, FSD could stop repeating some of the same choices that cause an owner to take over in the first place.

One thing that really frustrates me is when Tesla Self-Driving exits the carpool lane or express lane and then gets stuck in traffic.

As I see everyone else whizzing by in the far left lane, I get extremely annoyed knowing I should be there too.

Respect the settings please 🙏

— Whole Mars Catalog (@wholemars) July 18, 2026

The complaint that sparked Musk’s reply was wonderfully specific. The driver wanted the system to stay in the carpool or express lane, yet the car would leave it and get stuck in slower traffic.

That is exactly the kind of small, repeatable annoyance personalization should solve. A setting can tell the car what a driver generally wants; an intervention shows the precise moment the software failed to deliver it.

The car will start to remember your specific interventions and match each person’s individual preferences

— Elon Musk (@elonmusk) July 18, 2026

Musk’s sentence is brief, but “each person’s” is the important part. This sounds less like one fleet-wide behavior change and more like a layer that can distinguish how different drivers use the same car.

It also leaves big implementation questions unanswered. Musk did not say whether that memory will follow a Tesla account, stay with one vehicle, attach to a driver profile, live in the cloud, or require an owner to approve what the system learns.

Tesla Support describes the current product as Full Self-Driving (Supervised), and that last word still matters. The driver must remain attentive, keep control of the vehicle and be prepared to intervene because the system does not make the car autonomous.

Tesla already gives owners explicit ways to shape the experience, including selectable speed profiles and an HOV-lane control. Those settings establish a broad preference through explicit controls, but they remain fairly blunt.

Remembering interventions would be more granular. Enabling HOV lanes tells the car that it may use them; it does not capture every judgment the owner makes about when to stay put.

The software could potentially learn that this driver does not want to abandon a fast-moving express lane for congested general traffic.

Tesla’s Model 3 Owner’s Manual explains that an FSD speed profile affects driving behavior and lane changes, and that the selection is associated with the active driver profile. Depending on availability, the choices include Sloth, Chill, Standard, Hurry and Mad Max.

That is already a form of driver-specific customization, but it is still a menu choice. It tells FSD how assertively to operate across many situations rather than teaching it what one owner prefers at one recurring decision point.

The manual also says FSD uses navigation information and historical fleet data when planning a trip. Musk’s new description suggests a more personal input could join those broader signals: the history of what this driver corrects.

Not a Tesla App reported in April that software update 2026.2.9.9, paired with FSD 14.3.2, refined the feedback menu that appears after a disengagement. Drivers were given four reasons for taking over: Preference, Discomfort, Critical and Navigation.

The menu reportedly remains on screen until the driver selects a reason. That gives Tesla structured feedback capable of distinguishing a critical intervention from discomfort, a navigation problem or a simple preference.

That existing feedback path makes Musk’s answer especially interesting, but it does not prove the new behavior is live. Collecting a driver’s reason for intervening and changing future driving for that specific person are two different steps.

The best version of this idea would make FSD feel less like a talented stranger driving your car and more like a system that actually gets to know you. It could learn when you prefer to remain in an express lane, which route choices you routinely reject and where you want a calmer or more decisive move.

The guardrails will matter just as much as the learning. Owners should know what is being remembered, where it follows them, how to clear it and whether a mistaken intervention can be unlearned.

For now, Musk has described a direction, not a shipped feature. Even so, it points toward one of the most practical FSD upgrades Tesla could make: a car that drives the route and remembers how its owner wants that drive to feel.