The context
Realistic environments are hard to come by.
Gumloop's agents operate across 50+ SaaS tools. A workspace with three test issues doesn't exercise the edges those agents hit in production.
What we built
Live platforms with simulated activity from virtual teams.
One virtual team across every connected platform
Months of cross-platform activity, regenerated on demand
Dijon Data's virtual environments behave like our real customer setups… our agents test against the same patterns and quirks they'll see in production.Shreyas Prasad, Engineer, Gumloop
| Component | Spec |
|---|---|
| Personas + workflows | Each persona has a role, a department, and scoped access on each platform. Workflows span tools: a Linear issue, its Notion spec, the GitHub PR that ships it, and the Slack announcement all share the same task ID and owner. Reseed the environment and the references still line up. |
| Platforms | Agents self-onboard: read the platform's docs, sign themselves up, and start producing data with minimal supervision. |
| Dual-channel seeding | Agents use API calls and browser actions interchangeably, the same way a real team does. |
How it works
Synthetic org, real platforms, living data.
An org specification feeds work plans per platform. Agents act out each persona's plan via API calls and browser automation, producing environments that look like months of real company activity.
Virtual enterprise delivery / synthetic org → agent execution → live platform data / Dijon Data 2026
Outcome