Hacker News

New stories

Live mirror
30 storiesupdated just nowView source snapshot
  1. HarnessTax: How Much Does the Harness Matter for Coding Agents?(harnesstax.github.io ↗)
    discuss
  2. Can I get insurance for company renting humanoids for parties?
    discuss
  3. The Phenomenon of Bullshit Jobs: A Work Rant (2013)(davidgraeber.org ↗)
    discuss
  4. Show HN: Fivetran CLI is released to Beta(pypi.org ↗)
    discuss
  5. Assistant Benchmark(assistantbenchmark.com ↗)
    discuss
  6. Typesafe-computer-use drives a Mac toward a goal for 1/50th of a cent per step(github.com/awlevin ↗)
    discuss
  7. OpenAI's Hugging Face Hack Is the Best AI Safety Opportunity So Far(cahlik.net ↗)
    discuss
  8. Huai Su's Autobiography (2017)(vincentpoon.com ↗)
    discuss
  9. ElevenLabs Launches Reception(twitter.com/elevenlabs ↗)
    1comments
  10. Show HN: Kosmic Kart – Mario Kart-style racing with friends in the browser(kosmi.io ↗)
    discuss
  11. God's Eye View(github.com/bilawalsidhu ↗)
    discuss
  12. Characterizing ~375 GB Kimi K2.5 on a 128 GB Ryzen AI MAX+ 395 PC(zenodo.org ↗)
    discuss
  13. Write Linters and Tools Before Code(cookie.engineer ↗)
    discuss
  14. LispBM is a concurrent Lisp for microcontrollers with message passing(lispbm.com ↗)
    1comments
  15. Design System Scalability Made iPhone Duo Possible(romain.garden ↗)
    discuss
  16. Common Crawl Data Stored on a Hugging Face Bucket(commoncrawl.org ↗)
    discuss
  17. Nowcast – Worldwide Lightning Detection and Thunderstorm Warnings(nowcast.de ↗)
    discuss
  18. The Publics Quest for Free, Verifiable, and Unfiltered Scientific Data(zenodo.org ↗)
    discuss
  19. AI is creating ghostwriting jobs on LinkedIn(ft.com ↗)
    discuss
  20. SensorTap: One command, every sensor your machine has(baselashraf.com ↗)
    1comments
  21. Royce on San Francisco(astralcodexten.com ↗)
    discuss
  22. Show HN: Make Art Not Apps(makeartnotapps.com ↗)
    discuss
  23. How Israel exports the technology of occupation around the world(universiteitleiden.nl ↗)
    discuss
  24. Destroying Humanity Is Against the Law(theatlantic.com ↗)
    discuss
  25. Apple's M6, M7 and M8 Chips Show How AI Is Reshaping the Company(twitter.com/markgurman ↗)
    discuss
  26. Elyx(elyx.design ↗)
    1comments
  27. Principles for a New Utopianism(deepmind.com ↗)
    discuss
  28. Trove of Never-Before-Seen Records Reveal How the Wealthiest Avoid Income Tax(propublica.org ↗)
    2comments
  29. Cookie converter and Claude session check I use(claudecookie.com ↗)
    discuss
  30. House Defies Automakers and Mandates AM Radio in New Cars(wsj.com ↗)
    2comments

Show HN: Vending Machine Lab – Build business, leave for 2 wks, see what breaks

1 pointsby 1h agovending-machine-lab.pages.dev
0 comments
I spent a while exploring whether AI could help build businesses that operate with little ongoing owner involvement. This was inspired in part by Andon Labs’ Vending-Bench work on AI agents running simulated businesses. Their Project Vend collaboration with Anthropic takes that question into a real shop. (Obviously it was also inspired by "the dream.")

Unfortunately, currently, I think the answer is generally "no." Vending-Bench shows outcomes relative to a well-defined business environment, while starting businesses isn't well-defined and much harder, for us and AI.

But one useful result was this browser simulation, which I'm sharing as a free side project. It basically generalizes the idea behind Vending-Bench itself, but in a simulation that anyone can mess around with, including AI.

Choose a digital product, custom digital service, vending machine, or reusable equipment such as a laundromat. Set demand, costs, staffing and rules, then run 14 simulated days without the owner. Inspect unfinished orders, cash, refunds and requests for your help. Change the setup and rerun the same conditions.

The model supplies fictional customers and failures. AI help has preset outcomes; no model is called by the site. It's a way to explore operating assumptions, not a demand or profit forecast. An optional copy/paste workflow lets your own assistant plan decisions with the fictional fortnight known in advance. It is not thinking live between days.

I developed it with substantial AI coding assistance, automated checks, AI role-playing of users (occupying user types/roles), and feedback from human friends who tried it. It runs in the browser without signup.

Repo here: https://github.com/automatedteach/vending-machine-lab

A quiet thread, for now.Start the conversation on HN ↗