The goal wasn't just to pick a "driving playlist" at the start of the trip. Instead, I wanted to build an AI workflow for music selection that treats a journey as a living trajectory.
The technical logic behind Longplay #
Most music apps rely on static moods. Longplay is different because it functions as an LLM-adjacent agent for your Spotify queue, using live vehicle telemetry to drive transitions. It doesn't just look at genre; it looks at the state of the drive.
The system ingests several data points to mutate the queue dynamically:
Pace and Traffic: Adjusting tempo based on speed and congestion levels.ETA and Journey Phase: Shifting the energy as you approach a destination or enter a long stretch of cruising.Environmental Context: Regional data, light conditions, and even specific events like crossing a border or exiting a traffic jam.
The idea is to combat what researchers call "Driving Without Attention Mode." In this state, the operational mechanics of driving (staying in the lane, braking) continue on autopilot, while the driver's conscious attention drifts elsewhere.
The battle against passive fatigue #
When I was coding this, I kept thinking about the distinction between active fatigue (being overloaded) and passive fatigue (underload). Research shows that highly automated or monotonous conditions reduce task engagement and can actually slow down emergency responses.
This is where the "highway as pasture" concept comes in. On a long, empty stretch of road, the mind begins to graze. It wanders. Studies have even shown a correlation between highly distracting mind-wandering and crash responsibility.
My hypothesis was simple: If monotony causes attention to drift, can a seamless, context-aware setlist keep the driver "engaged" without becoming a cognitive distraction?
Real-world deployment and observations #
The trip was a chaotic, real-world deployment. It wasn't a controlled lab setting; it was a family trip with charging stops, overnight stays, and varying terrains.
A few things stood out during the long stretches:
Regional Transitions: Using music that reflects the geography made crossing borders feel like a distinct event rather than just more asphalt.Night Driving: The system shifted toward more relaxed, ambient textures that matched the lower visual stimulation of night driving.The "Feeling" Problem: I came home feeling convinced the system made the drive better. But as a dev, I had to check my bias. Is the system actually improving engagement, or did I just enjoy the novelty of my own code?
From a pure engineering perspective, the integration of telemetry into an LLM-driven or rule-based music agent is a fascinating way to bridge the gap between physical sensor data and subjective human emotion. We are moving toward a world where our AI agents won't just manage our calendars, but will actively tune our sensory environments based on our physiological and environmental states.
If you want to look at the implementation, the repo is here:
https://github.com/Hiepler/longplay
Next Stop treating AI like a magic wand and start acting like the →