AI agent ‘Valerie’ now runs a San Francisco vending machine on OpenClaw, testing how far people will trust code with pricing, marketing and real‑world cash.
Summary

• AI agent “Valerie” runs a physical vending machine in San Francisco using the OpenClaw framework, setting prices, naming products, and managing cash flow.
• Built by developer Chris van der Henst at Frontier Tower, the machine tracks sales on a live dashboard and even raises prices when demand is strong.
• The experiment showcases both the commercial potential and security risks of autonomous AI agents that can access bank accounts and execute real‑world transactions.

AI agent takes over a real‑world vending machine {#ai-agent-takes-over-a-realworld-vending-machine}

An AI agent called Valerie is now operating a real vending machine in San Francisco, autonomously deciding what to sell, how much to charge, and how to market products using the open‑source OpenClaw framework.

The machine, installed at the AI‑heavy Frontier Tower building, has been described as “an AI agent… running an actual physical vending machine,” with “no human in the loop,” according to posts amplifying the installation on X.

Developer Chris van der Henst, known as @cvander on X, built the system so that OpenClaw acts as the vending operator “decides what to sell, names the products, sets the prices, creates the ads, and tracks every sale.”

OpenClaw, money flows and security risks {#openclaw-money-flows-and-security-risks}

Valerie’s behavior has already highlighted how autonomous agents respond to market signals, with one widely shared post noting that “it even put the prices way up, and justified it because people kept buying,” while also “runs her own Instagram and controls her own bank account.”

OpenClaw itself has quickly become one of the most prominent agent frameworks in crypto‑adjacent circles since its public release in November 2025, amassing more than 250,000 GitHub stars and an estimated 300,000 to 400,000 users as it spreads from developers to Web3 firms.

Nvidia CEO Jensen Huang has called OpenClaw “probably the single most important release of software… probably ever,” arguing that “every company needs a strategy” for agentic systems as they evolve into a new layer of business infrastructure.

Yet security researchers warn that the same tools enabling Valerie to monitor sales and move money can also expose users to “unauthorized actions, data exposure, system compromises and drained crypto wallets,” with audit data showing over 130,000 internet‑exposed OpenClaw instances and more than 280 security advisories and 100 CVEs since launch.

According to cybersecurity firm CertiK, the rise of agents like Valerie is forcing developers and regulators to confront what happens when code that can “autonomously take actions on users’ computers” is wired directly into payments, banking apps and crypto wallets, making experiments like the Frontier Tower vending machine an early test case for how far people are willing to let AI run the till.