State Machines launches infrastructure to run thousands of enterprise environments in parallel

for agent testing A company called State Machines has released infrastructure that spins up thousands of isolated enterprise environments at once, each preserving its own state, so AI agents can be tested against the real API of the systems their customers use every day. The premise is straightforward: instead of building brittle mocks or begging for production access, developers get working copies of Salesforce, SAP HANA, Zendesk, Workday, ServiceNow and dozens of other services, ready to run in seconds.

How it works in practice

Each environment maintains state across the entire test run. An agent can create a lead in Salesforce, update the account, close the deal, open a ticket in Zendesk and post an invoice in SAP, and everything persists between calls. The platform supports a long list of integrations: the Google suite (Workspace, Gmail, Sheets, Drive, Docs, Calendar), developer tools (GitHub, GitLab, Linear, Jira, Confluence, Figma), financial systems (Stripe, QuickBooks, Xero, Sage, ADP, Gusto), communications (Slack, Discord, Zoom, Intercom, Mailchimp, Twilio, Resend), storage (Box, Dropbox, Notion), automation (Zapier, Asana), observability (Datadog, PagerDuty), data (Snowflake), identity (Okta) and commerce (Shopify).

Thousands of identical runs, not simulation

The model lets teams execute the same scenario a thousand times in parallel on identical environment copies, yielding a consistent score for the agent against realistic "customer worlds." This is not static simulation: the APIs respond like the real thing, with the same rate limits, the same errors and the same data shapes. For teams building agents that must operate inside a complex enterprise stack, it replaces months of internal test-infrastructure work.

What the announcement leaves out

The release does not disclose a pricing model, geographic availability, or whether environments run on State Machines' cloud or can be self-hosted. It is also unclear how sensitive data is handled when tests touch systems like ADP or Snowflake, and there is no mention of SLA or enterprise support. The company positions the product as ready to use, "ready to test in seconds," but offers no technical detail on architecture, environment versioning, or how updates are handled when Salesforce or another vendor ships a new release.

The bottom line

State Machines is selling developer time: instead of building and maintaining mocks for every integration, teams can start testing agents against the real thing from day one. The practical question is whether performance, cost and edge-case coverage live up to the promise, or whether this is another abstraction layer that breaks when an agent hits a genuine 429 from Stripe in the middle of a sales flow.