Pricing for coding agents is shifting from seat-based models to credit- and wallet-based systems as agents can execute end-to-end value on platforms, according to Andrew Garvin of Metronome (now part of Stripe).

Developer experience is framed as a safety mechanism, featuring skill files to guide agents, verbose and self-correcting errors, and sandbox-first testing to ensure human oversight before production.

The core message is that agentic commerce is active now, with success hinging on robust guardrails, effective metering via credits and commitments, and a human-in-the-loop approach to safety and governance.

A notable risk is runaway spend by agents, which wallet-based controls address by capping spending and enabling customer risk management alongside revenue metering.

Coding-agent firms such as Cognition, Cursor, OpenAI, and Anthropic are adopting cloud-service-style commitments with prepaid and postpaid credits to manage spending and commitments.

Stripe Projects and Metronome integration is presented as a template for making services discoverable to agents, with providers like Vercel and Hugging Face onboarding to the platform.

Three agent roles in commerce are identified: agent as product (pricing by usage), agent as buyer (provisioning services for businesses), and agent as user (operating platforms as end users), with the last role accelerating credits/commitment pricing.

A live demo demonstrated a coding agent provisioning Stripe Projects, Metronome, and backend services via a single natural-language prompt, highlighting potential while underscoring the need for guardrails and human review before production.

HubSpot’s shift from seat-based to credits-based pricing is highlighted as a leading indicator that a single agent can capture most value, rendering seat counts obsolete in some contexts.