Quick Overview
This product demonstration video presents Claude Code running the Fable 5.1 model with a one-million token context window. It illustrates an automated developer workflow where an AI agent diagnoses an intermittent vehicle climate issue across support tickets, telemetry data, and source code in the background.
Key Points
- 1.Claude Code running Fable 5.1 can investigate complex asynchronous system bugs autonomously across support tickets, telematic logs, and source code.
- 2.Analyzing telemetry across 214,312 events identified that 90.1 percent of delayed cabin heating requests experienced an exact 600-second delay.
- 3.The failure occurred because remote wake requests created a timing gap where the vehicle acknowledged commands before the thermal electronic control unit woke up.
- 4.Physical test vehicles never reproduced the bug because manual key unlocks wake both the car and the heater simultaneously.
- 5.Adjusting the application retry timer from 600 seconds to 90 seconds resolved the cold-start climate failure without requiring electronic control unit firmware updates.
Summary
The demonstration opens on vehicle ticket number 89289, illustrating a center display simulation where a cabin heat request set to 72 degrees Fahrenheit fails after repeated retries. The simulator interface displays an alert indicating that a climate request packet was acknowledged on the vehicle bus, but no status confirmation arrived from the thermal control unit. A developer opens Claude Code running Fable 5.1 with a one-million token context window, pastes the simulation logs noting that customer heat requests sometimes take up to twenty minutes or fail entirely, and instructs the agent to investigate while the developer shifts to another pull request.
Running in autonomous background mode, Claude initiates three explore agents to read support tickets, join telemetry from the electronic control unit, and extract application climate logs. The agent matches every command sent with the corresponding heater confirmation and measures the time gap between them. After stripping intermediate retries across 214,312 calculated deltas, Claude discovers a constant delay of 600 seconds, with histogram analysis revealing that 90.1 percent of delayed events fall between 598 and 612 seconds, 5.5 percent fall near 1200 seconds, and 4.4 percent expire entirely.
Claude hypothesizes that the 600-second delay represents an application retry timer masking underlying communication drops. When the engineer notes that the thermal electronic control unit cannot receive over-the-air updates, Claude searches the repository for the 600-second constant and locates a macro definition in the wake scheduler source file. Claude explains that after a recent wake update, heating activation became a two-step sequence where the vehicle wakes up before the heater module is ready, leaving early remote start requests unhandled until the 600-second application timer triggers a retry.
When asked to prove the diagnosis, Claude outlines that the heater module wakes within 30 to 46 seconds of the vehicle, meaning a 90-second retry interval provides double the worst-case wake window plus jitter. Physical test vehicles never exhibited the bug because enabling the door with a physical key wakes both subsystems simultaneously, unlike remote cold starts. The simulation is rerun using build 4.12.0-rc3 with the new 90-second retry logic, confirming that the cabin heating request succeeds and the vehicle interior heats smoothly from 48 degrees to 72 degrees Fahrenheit.
Reproducing the Vehicle Climate Failure
The demonstration begins with vehicle ticket number 89289, where a cabin climate request drops after five retries without registering an electronic control unit fault code. A developer pastes the customer simulation log into Claude Code running the Fable 5.1 model with a one-million token context window, instructing the agent to diagnose the intermittent heating failure in the background while the developer works on a separate pull request.
Multi-Agent Telemetry and Log Analysis
Claude launches three explore agents in parallel to ingest support tickets, join telemetry from the electronic control unit, and pull application climate logs. By aligning each heat request with heater activation timestamps across 214,312 data points, the agent discovers that 90.1 percent of delayed starts stall for exactly 600 seconds, pointing to an application-level retry timer rather than an onboard vehicle failure.
Isolating the Root Cause in the Wake Scheduler
Because the thermal electronic control unit cannot receive over-the-air updates, the fix must reside in the software application. Claude inspects the codebase to locate a 600-second retry definition in the wake scheduler, explains why key unlocks masked the issue during physical testing, and proposes shortening the retry timer to 90 seconds. A rerun of the simulator benchmark on build 4.12.0-rc3 confirms that cabin temperature successfully reaches the 72-degree target.
The Bottom Line
The video demonstrates how Claude Code running Fable 5.1 can independently isolate complex timing and telemetry bugs across disparate data sources without human intervention. It establishes that a persistent automotive heating delay was caused by a mismatched application retry timer during remote cold starts rather than hardware malfunction. The demonstration concludes with successful benchmark verification in the simulator, though deployment across the broader production fleet remains outside the scope of the video.
FAQ
What is Claude Fable 5.1 and how does it assist with debugging across the whole stack?
Claude Fable 5.1 is an AI model used within Claude Code featuring a one-million token context window, designed to run autonomous agents that read support tickets, analyze telematics, and modify source code to diagnose complex software bugs.
Why did remote vehicle cabin heating requests fail to activate the heater immediately in the simulator log?
A recent wake update split heat activation into a two-step sequence where the vehicle wakes before the heater control unit. Remote requests sent during that transitional window were acknowledged by the bus before the heater was listening, leaving the command unfulfilled until a later retry.
Why could the automotive climate failure not be reproduced on physical test vehicles during testing?
Physical test cars were enabled using a key, which wakes both the vehicle systems and the heater control unit simultaneously, eliminating the timing gap that only occurs during remote cold starts.
How did Claude Code analyze telematics data to isolate the 600-second delay in cabin heating requests?
Claude ran three background explore agents to join telemetry with app logs and customer tickets, calculating the time difference between heat requests and heater activation across 214,312 events to find that 90.1 percent of delays clustered at 600 seconds.
What code fix and retry timer adjustment resolved the vehicle cabin climate failure in build 4.12.0-rc3?
The application retry timer in the wake scheduler was reduced from 600 seconds to 90 seconds, providing double the worst-case 46-second heater wake time plus jitter, allowing the cabin to heat reliably within two minutes.
Worth watching for
Software engineers, systems architects, and embedded automotive developers interested in using AI coding agents for full-stack debugging and root-cause analysis.
- claude-code
- fable-5-1
- debugging
- telematics
- automotive-software