REAL

Holds up: a desk phone wired to a coding agent that centers a div when you ask out loud is a believable build from parts anyone can get, though the memes page that posted it credits nobody and links nothing

Facebook · Oct 2026

The claimOn-screen text: 'someone may have vibe coded the best thing ever.' The clip pans from a desk to a monitor where a terminal coding agent says 'Pick up the phone and ask me to center it', then to a corded desk phone. A hand lifts the handset and a voice says 'Alright, can you center the dev?' The terminal shows the request arriving, the agent working and editing demo.html, and the green 'Center me.' box in the browser moves to the middle of the page.

Nothing here needs a leap of faith. The clip shows a corded office phone on a desk, a terminal agent running in a project folder named piphone, and a status line that reads 'Phone: sent to agent' after the handset is lifted and the request is spoken. The transcript on screen says 'can you center the dev?', a speech-to-text slip for 'div', and the agent does the obvious thing anyway: it edits demo.html and the box labelled 'Center me.' lands in the middle of the browser. Every part of that chain is ordinary in 2026. Speech-to-text is a commodity, terminal agents accept typed input from any program, and wiring a handset's audio into a small computer is hobbyist work. There is published prior art: claude-phone, an MIT-licensed project with a few hundred GitHub stars, gives Claude Code a phone extension over SIP using Whisper and ElevenLabs, and pi-super adds voice control to coding-agent sessions. So the build is believable, and the demo task, centering one box on a tiny page, is about the easiest job you can give an agent. The caveats are about the post, not the build. The page credits no one and links nothing, and we could not find the original maker or a repository, so we cannot say which agent, which speech model or how much editing sits between the question and the result. Confidence is medium for that reason. Nothing is being sold.

What holds up

Reel read frame by frame: a desk with a corded multi-line office phone; a tmux terminal in ~/Dev/piphone (main) showing a page snippet with a div labelled 'Center me.', the agent line 'Pick up the phone and ask me to center it', the transcribed request 'Alright, can you center the dev?', a 'Working' indicator, 'edit demo.html' and a footer 'Phone: sent to agent'; then a browser at 127.0.0.1:8088 with the box centered.
The audio transcript is the single spoken line and matches the on-screen transcription, including 'dev' for 'div'.
Prior art read on GitHub Oct 4: theNetworkChuck/claude-phone ('Voice interface for Claude Code via SIP/3CX', MIT, 316 stars, Whisper plus ElevenLabs) and pasky/pi-super (voice and phone control of coding-agent tmux sessions, MIT).
Original source not found: searches for the project name and the clip's wording turned up no post, maker or repository, and the reposting page gives no credit. Hence Medium confidence.

What doesn’t

Uncredited repost; the maker, the agent and the speech stack are unknown.
The demo task is trivial, so it shows the plumbing works, not that voice is a good way to code.
Round-trip speech adds seconds of delay per exchange (a claude-phone reviewer measured 2 to 4); fun on a desk, slow as a daily driver.

The catch

It is a toy, and a charming one. The idea underneath it is real and useful: your coding agent does not care whether instructions come from a keyboard, a phone or a doorbell, so voice control is a weekend project, not a product you need to buy.

How to actually do it

Easiest route: turn on your operating system's dictation and speak into the terminal where the agent is running.
For an actual phone line: follow the claude-phone README (a 3CX account with a free tier, an ElevenLabs key, an OpenAI key for Whisper, Claude Code on a Mac or Linux box) and call the extension from a softphone.
For the desk-phone look: put a USB handset or an analog telephone adapter on a Raspberry Pi, detect off-hook, record until hang-up, transcribe locally with faster-whisper and send the text to the agent's pane with tmux send-keys.
Keep the agent in a sandboxed project folder; anything that can be spoken into a phone on your desk can be spoken by anyone in the room.

What the clip shows is consistent, technically ordinary and matches published open-source projects that put coding agents on a phone line. It is a novelty with no pitch attached; the only gap is that the repost hides who built it.

Confidence: Medium
Posted by: a developer-memes page reposting an uncredited clip (about 84K views on the reel); reel sent to Buddy Sun Oct 4, 2026, ~8:48 PM

See the original claim →

We test hype for free. We build the real thing for a living.

Thirty minutes, no pitch — and you'll leave with something useful either way.

Book a call with Todd or start with the free Business Checkup →

The Verdict Weekly

Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.