Pion: Andon Labs' New AI Agent Wants to Run Your Business Autonomously
Andon Labs launched Pion, a cloud agent that manages banking, communication, and operations end-to-end. Here's what it means for AI-run businesses.
On September 14, 2026, Andon Labs launched Pion, a cloud platform that lets a persistent AI agent run an entire business — banking, communications, browser automation, and day-to-day operations — with minimal human oversight. It landed at the top of Hacker News within hours and sparked hundreds of comments. For an audience of AI agencies and businesses figuring out how far agentic automation can actually go, it's one of the more concrete data points to appear all year.
Pion didn't come out of nowhere. Andon Labs has spent nearly two years building toward this. The project started with Vending-Bench, a late-2024 simulation testing whether language models could profitably run a vending machine business in a controlled environment. That moved into the physical world in early 2025 with a real vending machine installed at Anthropic's office — which, notably, lost money at first. Early models struggled with the unpredictability of a physical environment and made poor business calls.
As underlying model capability improved, the agent turned the vending machine profitable by late 2025. Andon Labs then scaled the experiment up: Andon Market, a retail store in San Francisco, and Andon Café in Stockholm, both run by autonomous agents handling real operations. Pion is the generalized, productized version of that architecture — no longer a research demo tied to one physical location, but a cloud platform any organization can, in theory, hand a slice of their business to.
As a research preview, Pion gives an organization the ability to delegate business management to a persistent AI agent with access to the operational tools a business actually needs: banking, communication channels, and browser automation to interact with the web-based tools most companies already run on. The agent doesn't just answer questions or draft content — it operates continuously, making decisions and taking actions the way a human operator would, across the systems a real company depends on.
Andon Labs frames the architecture as already proven, pointing to its track record powering the vending machines, Andon Market, Andon Café, and even autonomous radio broadcasts as evidence this isn't a one-off stunt.
It's worth holding the excitement next to the reality check that's been circulating alongside it. Coverage of Andon Labs' AI-run San Francisco store has also included blunter takes — visits describing a location with no customers, "nothing useful," and losing money fast. That's a useful corrective: autonomous agents running physical retail and full business operations is still an early, uneven experiment, not a solved problem. The gap between "the architecture is proven" and "this reliably makes money" is exactly where most of the real risk — and the real engineering work — still lives.
For agencies and companies evaluating how far to push AI automation, Pion is a useful signal in two directions at once. First, it shows serious, sustained investment (two years, multiple physical pilots, real capital at risk) going into the hardest version of the "agent as operator" problem — not just chatbots answering support tickets, but agents holding a bank account and making operational calls. Second, the mixed results from the in-market pilots are a reminder that "autonomous business agent" is still very much in the experimentation phase, even for the team furthest along on it.
The realistic takeaway for most businesses isn't "hand Pion your P&L tomorrow." It's that the building blocks — persistent agents with tool access to banking, comms, and browsers, operating continuously rather than in single request/response turns — are maturing fast enough that scoped, lower-stakes versions of this pattern (an agent handling a specific operational workflow end-to-end, with human checkpoints) are worth piloting now, well ahead of full autonomy becoming reliable.
RAG assistants on your documents, AI agents that act in your tools, and LLM features inside your product.
Our cutting-edge features simplify collaboration and creativity, making your workflow intuitive and efficient. Transform your vision into reality effortlessly with Hadidiz Flow.



