Arga Labs has raised $10 million in seed funding to build resettable, stateful replicas of enterprise applications that let teams test AI agents across complex, multi-step workflows without touching live systems. The round was led by General Catalyst, with participation from Box Group, Emergence, Gradient and SV Angel; the financing was recorded in Dealroom’s August 26, 2026 funding report. The San Francisco startup was founded by Phillip Li and Akira Tong.
Arga Labs’ approach
Arga is developing isolated digital replicas of business applications that preserve evolving application state, permissions and webhooks as an agent moves through a workflow. Rather than returning isolated, stateless API responses, the replicas are designed to retain authentication, authorization, mutable resources, timing, webhooks, failures and retries so an agent can be evaluated on the cumulative effects of a chain of actions. The company cites applications such as Salesforce and Workday in descriptions of the platform.
Teams can prepare users, permissions, records and other shared state, connect an agent to the isolated environment, run a workflow and record what happened. Tests capture requests and responses alongside state changes, blocked actions, latency and side effects. After evaluating results against defined success or safety conditions, the environment can be reset to a known baseline so subsequent runs start from the same state.
Testing stateful workflows and failure modes
Arga targets engineering organisations building agents that act inside third-party business software rather than teams focused only on text generation. The testing model is aimed at agents that must remember changes and coordinate across communications, customer records, billing, development tools and other systems in which one operation alters what is possible next. That broader scope surfaces failure modes such as repeated operations, tenant-boundary crossings, expired credentials and conflicting states between connected applications.
A controlled replica can also be used to introduce conditions like denied requests, retries, timeouts and partial completions repeatedly, instead of waiting for such failures to occur unpredictably in production. Isolation keeps experimental changes away from production accounts, customer records, messages and payment activity, and a repeatable starting point supports regression testing, evaluation and reinforcement-learning workloads.
Fidelity and practical limits
The product and funding update says the application twins preserve authentication, authorization, permissions, mutable resources, webhooks, timing, failures and retries and makes a broader claim around backend behavioral fidelity. That claim leaves practical questions for prospective customers about how closely a replica matches the live service.
Coverage can vary by endpoints, resource types or command-line and management tools required by a particular workflow. Permission behaviour must reflect roles, tenant boundaries, delegated access and expired credentials to reproduce real operational conditions. Failure behaviour also matters: useful replicas need to reproduce partial success, duplicate delivery, eventual consistency, timeouts and provider-specific errors rather than offering only clean responses. Rate limits, quotas and throttling are additional challenges, since training an agent in an environment without realistic operational constraints could produce behaviour that fails when those limits apply in production.
The $10 million seed round backs Arga’s bet that stateful replicas can make agent failures observable, repeatable and resettable. What remains to be demonstrated is how comprehensively each replica mirrors its corresponding live service and how reliably success inside the replica predicts success in production. Versioned coverage information, detailed permission and failure testing and controlled comparisons between replicas and production would make those fidelity claims easier to evaluate.
