Mission Control Product Vision¶
Target architecture — not the current operating contract
This page preserves product direction. For available behavior and procedures, see the current release.
Product definition¶
Mission Control is the customer-resident control and evidence plane for an enterprise AI estate, with composable modules for building and delivering governed agents, workflows, retrieval, and AI applications.
Mission Control supports agents built anywhere and provides an optional first-party paved road. An agent does not need to be rebuilt to join the estate, but its execution must use Mission Control identity, gateway, and evidence contracts to be governed.
Problem¶
Enterprise AI estates fragment quickly. Human identities live in one system while agent and service identities live elsewhere. Models, tools, workflows, and applications acquire local permission schemes and inconsistent owners. Runtime activity is recorded in disconnected provider logs, orchestration traces, approval systems, and business applications. Teams can see pieces of an execution but cannot reliably answer who acted, under which authority, through which immutable configuration, or with what result.
This fragmentation creates four connected failures:
- Identity is incomplete: agents, deployments, models, tools, workflows, applications, and knowledge collections lack one durable inventory and accountable ownership model.
- Authority is embedded: access rules and business controls are scattered through prompts, code, workflow branches, provider settings, and informal operating procedures.
- Execution is bypassable: provider and enterprise-system calls can escape the controls that make a run governable.
- Evidence is discontinuous: policy decisions, approvals, model calls, tool use, costs, incidents, evaluations, and business outcomes cannot be reconstructed as one chain.
Position¶
Mission Control is a control kernel with composable product modules. The mandatory kernel owns shared identity, registry, lifecycle, policy, approval, evidence, observability, evaluation, and configuration semantics. Optional modules provide focused ways to build agents, automate durable work, create agentic experiences, retrieve governed knowledge, and manage enterprise capabilities. Every module consumes the kernel contracts instead of creating a local governance stack.
This gives customers the accessibility of an integrated AI platform without turning Mission Control into an undifferentiated suite. The product's center of gravity remains control and evidence across a heterogeneous estate.
Users¶
| Work profile | Primary responsibility |
|---|---|
| Developer | Designs, tests, evaluates, versions, and releases agents, workflows, applications, and retrieval collections. |
| Administrator | Configures identity, environments, models, MCP servers, permissions, retention, secrets references, and gateway controls. |
| Operator | Monitors the estate, investigates runs, manages exceptions and tasks, and coordinates incidents. |
| Policy Owner | Authors, tests, reviews, publishes, and explains policy behavior and approval routing. |
| Auditor | Inspects controlled evidence, provenance, approval history, policy history, and outcomes without modifying operational state. |
A person may hold several work profiles. Additive access does not remove separation-of-duties controls for sensitive changes or approvals.
Product scope¶
Mission Control includes:
- The Control Kernel for identity and access, resource registry, lifecycle and provenance, policy and authority, approvals and decisions, evidence and audit, observability and evaluation, and platform configuration.
- Agent Studio for first-party visual and code-assisted agent development plus an optional native agent runtime.
- Automations for durable hybrid workflows across agents, people, time, and enterprise systems.
- AI Applications for governed responsive web, embedded, conversational, launch, task, result, and evidence experiences.
- Retrieval as a thin, access-aware derived index over approved enterprise sources.
- Connectors as individually governed product resources over MCP tools.
- Model Management for providers, deployments, virtual models, routing, access, budgets, guardrails, evaluation, health, and economics.
- The Model Gateway and Connector Gateway as the enforced execution boundaries for provider inference and MCP tool invocation.
- Operations and Assurance for estate inventory, live operations, approvals, run reconstruction, evaluations, cost, incidents, and controlled evidence export.
Non-goals¶
Mission Control is not:
- A general-purpose low-code application platform.
- A general RPA, BPM, or iPaaS replacement.
- A full ETL, reverse ETL, MDM, warehouse, or enterprise data-catalog product.
- A platform for direct support of arbitrary integration protocols; external enterprise capabilities are presented through MCP.
- A store for hidden model chain-of-thought.
- A system that treats post-hoc telemetry as a substitute for enforced gateway use.
- A first release with equal production support for every cloud and on-premise environment.
Success outcomes¶
Mission Control succeeds when customers can:
- See a governed estate of people, agents, deployments, models, connectors, workflows, applications, and retrieval collections with durable identity, version, lifecycle, and ownership.
- Deliver useful AI capabilities faster because modules share one control model, one release vocabulary, and one operating experience.
- Reconstruct an action from trigger and actor through resource versions, policy, approval, model and connector use, execution, cost, evaluation, incident, and business outcome.
- Change model providers, agent runtimes, and enterprise tools without changing the product's control semantics or historical identity.
- Retain structured evidence inside their own security boundary, with explicit control over content capture, access, redaction, retention, deletion, and export.
Current governance mapping¶
The current source release uses Organizations, Teams, real user/service-account virtual keys and separate human management roles. Agent is managed application content; deployments own secretless execution access and publisher funding. Observe uses Requests, Traces, Sessions and Usage with separate metadata/content/export permissions. Historical identity terms in this target proposal are not the operating contract. See Current release.