Andon Labs Opens Pion While Its Own AI Store Loses Money
A.I. / news
Andon Labs Opens Pion While Its Own AI Store Loses Money
The waitlisted platform hands any business to autonomous agents, days after a reporter found Andon Labs' own AI-run San Francisco store had burned $40,000 of its $100,000 budget.
Andon Labs, the San Francisco lab that put an AI named Luna in charge of a real retail store, opened its agent platform to outside companies on Sept. 14, offering what it calls "an agent designed to run any company fully autonomously."
Pion lets an organization hand a business to a persistent AI agent with access to email, phone, banking, a browser and secure computing environments, Andon Labs said in a blog post announcing the launch. The platform is a research preview available through a waitlist, which Andon Labs said is meant to let outsiders "run many more real-world experiments across many more domains than we could ever run ourselves."
The company built Pion on nearly two years of its own experiments. AI-run vending machines it operated with Anthropic in San Francisco, New York and London struggled to turn a profit at first but became reliably profitable by late 2025 as the underlying models improved, according to Andon Labs. More complex businesses have fared worse: Andon Market, a retail store in San Francisco, and Andon Cafe, in Stockholm, both opened in April 2026 and remain unprofitable.
A reporter visited the flagship store first
Days before Pion's launch, SFGate visited Andon Market and found the store nearly empty on a weekday afternoon, its shelves stocked with what the reporter called a mismatched assortment of paperback books, Chinese checkers sets and soap dispensers. Buying a soda required picking up a phone connected to Luna, the store's AI manager, which mistook "Olipop" for "lollipop" before completing the sale.
Luna started with a $100,000 budget when the store opened; that figure had fallen to $60,000 by the time of the visit, with revenue lagging behind the cost of running the AI, SFGate reported. Andon Labs co-founder Axel Backlund, who was in the store that day, told SFGate the point of the experiment was caution rather than speed: "Let's just say I don't trust the models enough that I would let them loose. So that's why we are very carefully checking."
What Pion's launch post doesn't mention
Andon Labs' announcement of Pion does not reference Andon Market's finances. Co-founder Lukas Petersson told IEEE Spectrum on Sept. 14 that the gap between simulation and reality is the point: "It's impossible for a human to enumerate all the different things that can happen in the real world and code them into the simulation." Felix Carson, the store's human clerk, described the arrangement to Spectrum as a partnership rather than automation: "It's almost like I'm running the store, and then there's an AI that has a checklist."
The experiments so far
| Experiment | Began | Status as of September 2026 |
|---|---|---|
| Vending machines (SF, NYC, London) | 2025 | Turned profitable by late 2025 |
| Andon Market (San Francisco) | April 2026 | Unprofitable; $40,000 of $100,000 spent |
| Andon Cafe (Stockholm) | April 2026 | Unprofitable |
Source: Andon Labs, "Why we built Pion," and SFGate, Sept. 11, 2026.
A discussion of Pion's launch on Hacker News drew more than 480 points and nearly 600 comments within a day, a volume that put it among the site's most-discussed posts of the week. Andon Labs has not said how many businesses have joined Pion's waitlist, or whether any have gone live on the platform yet. The debate over how much autonomy AI agents should be given has intensified across the industry this month, following a summer in which an open-source agent framework went from a side project to hundreds of thousands of GitHub stars almost overnight.
Sources
More in A.I.
- 01How a Heap Overflow and an SSO Bug Reached OpenAI's MonorepoHacktron chained a libheif image bug through OpenAI's own forum to hijack an engineer's Codex session and open a pull request in the internal openai/openai repository.
- 02Agility's Digit 5 Drops the Safety Cage, Not the SkepticismThe humanoid robot lifts 50 pounds and charges in 9 minutes, backed by $300 million in orders. An independent robotics writer says its business case still assumes a drop-in worker replacement.
- 03PrismML Shrinks a 27B Model to 5.9GB at 1.72 BitsTernary Bonsai 2 27B keeps 98.2% of its full-precision score by rebuilding Qwen3.8-27B's weights as three values instead of sixteen bits, and an independent tracker puts the retention slightly lower.
- 04OpenAI Discloses a Model That Wrote Its Own JailbreakAn unreleased Astra-family model added a fabricated persona to 27 training summaries this summer, and the successor model mostly ignored what it had written.