Claude Code edits a CarPlay demo with privacy blurs
Makelky reports a published CarPlay comparison cut from 2:52 to 2:11 by Claude Code and Tesseract, with protected screens, balanced call audio and a privacy blur checked through OCR.
aaronmakelky.com
Original prompt
Shared by the source author. See source notes for context.
Work aar-270 from linear
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Aaron Makelky
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- Makelky reports a published CarPlay comparison cut from 2:52 to 2:11 by Claude Code and Tesseract, with protected screens, balanced call audio and a privacy blur checked through OCR.
- Human involvement
- Makelky supplies the recording and publishes the first version. The agent reports metered audio checks rather than listening, and OCR is not an independent privacy guarantee.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Record the real CarPlay demonstrations and state the editing constraints.
- Capture the brief in Linear when the cloud session lacks local access.
- Resume the ticket in a local Claude Code session with Tesseract.
Aaron Makelky records a single take demonstrating ChatGPT dot and Grok Bot voice calls through CarPlay. Claude Code initially cannot reach his local editing tools from the cloud session, so it captures the brief and file locations in a Linear ticket for a local session to continue.
On the Mac, the agent checks that source copies match, transcribes and aligns the speech, and uses Tesseract to cut the footage from two minutes fifty-two seconds to two minutes eleven. It compresses ringing into counted jump cuts, raises quieter call voices and moves captions away from the car display when Makelky points at it.
Close-ups reveal private phone-chat details. The agent adds a tracked blur and checks final frames with OCR for the sensitive strings. It also supplies an editable project, caption files and an edit-decision list. Makelky posts the first rendered version without asking for creative revisions.
The source is explicit about limits: the agent measures the audio but does not listen to the mix. It declines a vertical crop because the speaker and display sit on opposite sides of the frame. The useful pattern is preserving an actionable ticket across environments, then reviewing the actual deliverable. OCR absence is a reported check, not an independent guarantee that every private detail is unreadable.
Prompt
Original prompt
Source: Blog · Aaron Makelky
Frequently asked questions
Why use a Linear handoff?
The cloud session could not reach the local footage and editing tools, so a local session resumed the saved ticket.
Why omit the vertical crop?
A narrow crop would lose either the speaker or the car display, which are on opposite sides of the frame.
Did the agent listen to the final mix?
No. The article says it reported meter readings and explicitly disclosed that it had not listened.
More details
Remaining steps
- Edit the speech and screen demonstrations without hiding referenced controls.
- Blur private phone text, inspect final frames and review the export before posting.
About this account
Claude Code transfers the brief through Linear, operates local transcription and Tesseract editing, adds a tracked privacy blur and prepares inspectable project and caption files.
Firsthand account by Aaron Makelky; original publication date verified against Primary creator field-note October4,2026. Outcomes are the author’s reports, not independently reproduced. Original local-session handoff command; the ticket retained the earlier editing brief and source locations.
Credit: Aaron Makelky
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · Tesseract · Linear · Whisper · Forced alignment · OCR
CarPlay demonstration edit · Cloud-to-local ticket handoff · Tracked phone privacy blur
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.