← Back to library

Write a new driver for a printer your OS dropped

revolvingthrow, a Hacker News commenter who reverse-engineers most of their USB gadgets with AI help, had Claude write fresh drivers for an old printer a macOS update had stranded. Claude steered the session and asked for observations: did grayscale fade line by line, did the two pages differ in quality, what error showed with no paper loaded.

  • OutcomeHave a coding agent reverse-engineer an old USB device and write a driver, while you act as its eyes on test prints.
  • SetupAn afternoon+
  • LevelAdvanced
  • Creditrevolvingthrow on Hacker News
View original
Works with
Claude Code

Published Oct 23, 2026Updated Oct 23, 2026AI dry run on sample input: passed Oct 23, 2026Source verified Oct 23, 2026

When should you use it?

  • A device lost support on your OS
  • You are comfortable with a terminal
  • You can run test prints or tests

Not for: Phones and secured devices; Devices under warranty you may void

What do you need, and what do you get?

You need

  • Device connected by USB
  • Test materials such as paper
  • Notes on each test result

You get

  • Working driver
  • Error handling for common faults
  • Test log

How does it work, step by step?

  1. Connect and probePlug the device in and let the agent read what it reports.
  2. Start simpleGet a basic test working first.
  3. Be its eyesDescribe each result precisely, as the agent asks.
  4. Add error handlingTrigger common faults such as no paper and let it handle them.
  5. Test featuresCheck quality settings and duplex last.

The prompt

Our own prompt, inspired by revolvingthrow on Hacker News and written to work in Claude Code. Values in lime are placeholders.

Fill in the blanks (optional; the copied prompt uses your values)

Prompt2 variables
I have a {device} connected by USB that no longer works with {os}. Probe what it reports and propose a plan for a driver.
Ask me to run one test at a time and tell you exactly what I see. Do not send commands that change firmware. Keep a log of each test and result.

Inspired by revolvingthrow on Hacker News: Comment in Using AI to reverse engineer hardware. The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.

What can go wrong?

  • Sending bad commands can jam or damage hardware; go slowly.
  • Keep notes so you can repeat a working setup.
  • Secured devices are off limits.

Our take

revolvingthrow's printer story flips the usual roles in a useful way. The model planned each test and the person became its eyes, reporting what came out of the tray. For hardware the agent cannot observe, that division works well, as long as your reports are precise: which line faded, which page looked coarser, the exact text of an error. Vague feedback such as 'it looks wrong' sends the agent guessing. Start with the simplest output, a single line of text, before graphics, resolution modes or duplex, and add error handling for paper-out and jams once printing works, as happened here. Be careful with commands the agent invents for undocumented devices. Printers usually shrug off bad input, but firmware updates, calibration routines and anything writing to device memory can brick hardware, so tell the agent those are off limits unless you agree. Keep the driver and notes in a small repository; the next OS update will probably break it again, and the fix will be much faster with history in hand.

Where the time goes back: An old device keeps working instead of going to landfill.

FAQ

Do I need hardware knowledge?
The commenter says they had little, but were comfortable with computers.
What was the human's role?
Reporting what each print looked like so the agent could debug.

Source and credit

Inspired by revolvingthrow on Hacker News. We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.

Inspired by
Comment in Using AI to reverse engineer hardware
Creator
revolvingthrow on Hacker News
Published on
Hacker News
Original date
Oct 15, 2026
License
Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
Originally for
Claude Code
AI dry run
Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. PASS: Gates held on desk check; no live sends, bookings or purchases.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails.
Verified
Oct 23, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.

Are you the creator? Claim or correct this page.

More in Career & Learning

All 37 Career & Learning playbooks →

Related playbooks

Career & LearningWorkflow✓ AI dry run

Have a coding agent review your Logseq journal

Point a coding agent at your note app's command line tool so it can summarize months of journal pages and report on tagged items such as job applications.

Credit: wosk on Hacker News

Under an hour
Career & LearningSkill✓ AI dry run

Learn any topic with a strict Socratic tutor skill

Load Jane Street's open-source tutoring skill into Claude, ChatGPT or a coding agent so that when you ask to be taught something, it first tests what you already know, keeps a 0 to 5 level for each part of the topic, quizzes you one question at a time without handing over answers, re-tests older material by surprise, and only ends the course after a final mixed exam.

Credit: janestreet

Minutes
Career & LearningWorkflow✓ AI dry run

Let a terminal agent hunt roles and fill applications

Turn your résumé into a job shortlist and submitted applications with matching cover letters, while unsuitable roles are skipped.

Credit: Artur Erofeev

An afternoon+