EXPLORINEER
en

UK — The Register

Vintage fun: Prodding AI agents using retro hardware just because you can

AI1 min read

A developer hooked up an old teletype printer to a modern AI coding agent, called Kiro. It is a fun, pointless-but-educational project that mixes 1970s-80s hardware with today's AI tools.

What happened

AWS developer advocate Darko Mesaroš connected a DECwriter teletype to AWS Kiro, an AI coding service that works from a command line. He can type to the AI and read its replies on paper, like an old-style computer terminal. He built a small Rust program to make the AI's output fit the old machine.

Key facts

  • The teletype prints about 30 characters per second.
  • A WiModem232 device connects the teletype using the Hayes modem standard.
  • The project uses Agent Client Protocol (ACP) to talk to Kiro.
  • The teletype's heyday ended about 40 years ago.

How it works

The teletype needs a serial connection, which is an old way of sending data one bit at a time. The WiModem232 acts like an old Hayes-compatible modem, a device that was once used to connect home computers to online bulletin boards. Kiro's normal output is too busy for a slow printer, so a Rust program trims it into plain text that the teletype can handle.

Why it matters

The project is more of a hobby stunt than a useful tool. Still, it shows that AI tools can work with many kinds of software when they follow shared standards like ACP. The builder says the same setup could work with other coding assistants, such as Claude Code or Codex. It also shows how much skill a project like this needs, including coding, containers, and protocols.

Things to know

  • The setup does not handle lost data packets. If data is dropped, it is gone.
  • There is no buffering, so the slow serial device is the bottleneck.
  • The code is public and built for Kiro, but the builder says it can be adapted.

Glossary

AWS Kiro
An AI coding service from Amazon Web Services that helps people write software.
Hayes modem
A common way of controlling old modems, used to link computers over phone lines.
Agent Client Protocol (ACP)
A shared set of rules that lets a program talk to an AI coding agent.
Rust
A programming language known for speed and safety.

Same story, different eyes

Ad

Add 150 words. Learn 10 a day.

Dicter word books send only the words you learn to your review notes.

AIVnExpress

Musk mocks Altman, says everyone deserves RAM too

Elon Musk mocked OpenAI CEO Sam Altman's remark that it is "bad" for users to run AI locally on their devices instead of reserving RAM for data centers. On November 2, Musk posted "Everyone deserves to own RAM" along with a meme featuring Christoph Waltz from Inglourious Basterds, and he made other jokes at Altman's expense. Altman did not respond to the posts.

AIZDNet Korea

Hanmi Pharmaceutical Builds Domestic Chatbot Infrastructure for Digital Healthcare Platform 'eppiday'

Hanmi Pharmaceutical has signed a strategic MOU with FuriosaAI, Xperix, and its healthcare subsidiary Baegle Labs to move the AI chatbot of its digital healthcare platform 'eppiday' to a Korean LLM and domestic AI infrastructure. Under the agreement, Baegle Labs will handle app and chatbot development, Xperix will provide domestic AI servers and LLM inference, and FuriosaAI will supply domestic AI chips. The four companies have verified in a test environment that the AI coaching answer generation can run on domestic servers, and the commercial launch timing will be set after quality and safety checks.

AIVnExpress

Anthropic Manager Drops Coding Tests in Hiring as AI Outperforms Humans

Felix Rieseberg, Anthropic's recruiting manager, says the company has stopped using live coding tests in hiring because its Claude models now write code better than humans. Instead, he is looking at candidates' portfolios for taste, empathy and principled thinking, and values those who build prototypes with AI. Surveys cited in the article show AI is making technical assessment harder for employers, with system design, debugging and AI collaboration gaining weight.

AITechNode

SenseTime open-sources 9-billion-parameter AI model that reuses layers for deeper processing

SenseTime has released SenseNova Loop-9B-Preview, an open-source multimodal model announced October 30 that reuses computation to extend processing depth without duplicating weights. It accepts text and images and generates text for document understanding, coding, and tool-use tasks, running a shared block of 24 decoder layers twice from a model of about nine billion parameters. SenseTime has published the weights, inference instructions, and training guidance under the Apache 2.0 license.