Agent traffic is rewriting how software gets built

For the first time, automated agents now account for more internet traffic than humans, according to Cloudflare. Stripe's own telemetry points to a specific driver: agents that write code. Traffic to Stripe's documentation grew more than 10x in 2025, with agents responsible for nearly 40% of it. The stripe-cli has seen exponential user growth over the same period, and agents now generate 70% of CLI requests for API resources.

Agents can clearly write and integrate code against APIs like Stripe's. The friction has shifted to adjacent tasks: provisioning infrastructure, managing credentials, setting up accounts, and wiring services together. Stripe Projects, which provisions and manages services from the CLI, is being expanded to close that gap with new agent integrations, additional providers, and finer-grained developer controls.

Agent-native provisioning arrives in three new tools

Stripe Projects is now available as a skill in Hermes, the open-source agent from Nous Research. A skill provides structured instructions that give the agent the context it needs to act. In this setup, Hermes maintains context across sessions while Projects handles infrastructure provisioning and management, letting an agent behave as a persistent collaborator on multi-day or multi-week product work.

Two other coding agents have also integrated Projects. Factory Droids and Warp, both model-agnostic, now expose the embedded Projects CLI directly within their workflows.

Provider catalog expands to 49

Projects has added 16 providers, bringing the total to 49. Notable additions:

  • Metronome (a Stripe product) for usage-based billing
  • Wix for public-facing storefronts
  • ClickHouse for LLM observability

With these, an agent can stand up a live, AI-powered product, meter and bill users, and monitor model cost, latency, and quality from the very first call—no human dashboard clicks required.

Guardrails for autonomous builds

Agents that provision infrastructure need the same controls as agents that spend money: scoped credentials, spend thresholds, and strict environment boundaries. Three updates address those requirements.

Unified spend visibility

Projects now shows cost per project across all providers, both current and historical. That gives a clear picture of what any given project costs to operate.

Per-provider spend caps

Instead of one blanket budget, you can set custom spending limits per service. For example, cap an AI model provider tightly while allowing a higher ceiling for production hosting or a database layer. This is designed to stop a single runaway provider from exhausting the entire budget.

Named, isolated environments

You can now mint isolated credentials for development, staging, production, or any custom environment, and choose to share resources across them where consistency matters. Agents default to the development environment, so an off-script agent cannot touch production data or infrastructure.

Platforms can provision on behalf of developers

With delegated authority and white-labeling support, platforms can now issue scoped credentials and set up infrastructure on their developers' behalf. The goal is to let a developer go from writing code to running it without ever leaving the platform environment.

Roadmap

Future Projects releases aim to cover the full lifecycle of agent-built software, from initial provisioning to ongoing operations and security. Planned work includes deeper security primitives for autonomous agents and a data layer so providers can meter and bill for agents that build on their services. Developers ready to try it can ask their agent to use Stripe Projects to add a Prisma database.