Andrew Garvin demonstrates combining Stripe Projects with Metronome to provision a test billing environment from a natural-language request. The example asks a coding agent to reproduce a Lovable-style credit model, including recurring credits, different usage categories and charges after credits are exhausted.
Andrew Garvin explains that detailed skills filesAn AI agent skill is a reusable package of instructions, resources, and tool guidance for performing a bounded kind of work. give the agent context about a complex billing API, while clear error messages help it recover from setup mistakes. The demonstration includes initialization problems, reinforcing the need to inspect what the agent has configured rather than assuming a natural-language request succeeded.
Andrew Garvin keeps the workflow in a sandbox and uses synthetic usageEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria. to exercise the billing configuration. He inspects the resulting customer, credit balance and draft invoiceAgent completion verification checks evidence that an agent achieved the requested outcome rather than merely producing intermediate work. to show how usage would consume credits and generate charges. He presents this as a way to accelerate testing, with people still responsibleHuman judgment in AI is the accountable interpretation and decision-making people contribute when setting goals, evaluating evidence, and managing consequences. for business logic and the production decision.
Andrew Garvin distinguishes agents as products, buyers of services and users of existing software. Each role creates different pricing and operational demands, from measuring consumption to controlling spending. He argues that usage-based and credit-based billing becomes more relevant when one agent can perform work previously spread across many licensed users.
Watch on YouTube
