Dome Systems

Patterns of AI agents

Standardize Agentic Apps

An assistant, a scheduled job, a developer's editor, a service for other people's agents. Each is a golden path your platform team sets up once and every app team reuses.

Who starts the call?A personA schedule or eventSomeone else's agentIn an app many people shareFrom their own editorInteractive AssistantMany people, one agentDeveloper HarnessYour own accountScheduled JobNo one watchingExternal ServiceOther people's agents

Example patterns

Standardizing agents with composable resources

Agents are typically sophisticated workflows, and so have similar patterns to produce expansive results. This enables platform teams to define golden templates and pathways for agents in production.

Hundreds of people, one agent.

A support copilot or an internal Q&A service. Each person expects answers scoped to what they can see.

CallersAgents · supervisingAgents · calledGatewaysTools · PoolsRulesRulesRulesGuardsCallersjane@OIDC verifiedCallersmarcus@OIDC verifiedCallershannah@OIDC verifiedAgentsconciergeacts-as requiredAgentsbillingAgentsknowledgeGatewayssupport-prodgranted per agentToolszendeskMCPToolscrmRESTPoolsprod-models3 membersQuotas
concierge → crm/get_account · as hannah@Allowed · Redacted

Hannah's identity travels with every call, through the sub-agent hop. A Rule can name her, and the Audit record names her and the agent. Click any box or gate to see what it is.

What makes it hard

  • One agent serving everyone reaches everything anyone can reach.
  • A sub-agent hop usually drops the person who asked.

The golden path

  1. Callers. Verify every person through your identity provider.

  2. Agents. Register the assistant with acts-as required.

  3. Gateways. Grant one Gateway of tools and model pools.

  4. Rules. Write conditions on the person, not only the agent.

  5. Guards. Redact what this person shouldn't see.

  6. Quotas. Cap the agent's spend over a window.

Next steps

Talk with our FDE team

Our forward deployed engineers work with your platform team to get your agents into production and under control: the first one governed on your own systems, and a pattern your teams can repeat for every agent after it.