Products / Control Plane

Many agents.
One organization.

Coordinate humans, agents, and organizational work.

One agent can perform a task. An organization needs people and agents to share context, ownership, approvals, and responsibility across teams and tools.

False Systems Control Plane is the organizational harness: a common coordination and authority layer across the agent tools teams choose.

Discuss shared agent work ↗
01 / Keep your agent tools

An agent harness
is only one layer.

An agent harness manages how an agent works: model interactions, tools, context, and execution loops. An organizational harness manages how agents and humans work together.

Teams may use Codex, Claude Code, Pi, or other harnesses and models. The design is to connect their work, not require a replacement agent runtime. These are examples of the environments the architecture addresses, not a list of shipped integrations.

Control Plane is a logical coordination layer. Local supervision and session lifecycle belong to Client; local actions do not each need a synchronous trip through shared coordination.

Organization

Humans · Teams · Policies

Control Plane / Organizational harness

Shared work · Ownership · Delegation · Approvals · Handoffs

Coding agentCodex harness
Testing agentClaude Code harness
Investigating agentPi harness
HumanReview & approval

HabitatOperations · Workers · Hosts

EvidenceActors · Grants · Observations · Outcomes

Conceptual relationships, not verified integrations with the named harnesses. People and agents share organizational context; local execution does not pass through a central service for every action.
02 / From individual tasks to shared work

Know who owns it.
Know what comes next.

Shared work
Relate the request, owner, progress, and dependencies so people and agents can work from a common picture.
Coordination
Make overlapping work and conflicting changes visible for a decision. Humans can intervene, redirect, or take responsibility; automatic conflict resolution is not assumed.
Delegation
Relate authority to a specific piece of work. A prompt or an unrestricted agent identity is not an authorization model.
Handoffs
Keep decisions and responsibility connected when work moves between people, agents, environments, or sessions.
Human decisions
Keep approvals explicit, with a person accountable for the privileged decision and its scope.
Accountability
Connect actors, delegated work, grants, and supporting evidence, including what has not been confirmed.
03 / A concrete example

Four conversations.
One related activity.

A developer requests a change. Other agents test and investigate. A human approves the production step.

Illustrative workflow / Not a live connected demonstration

  1. A human requests a service change.

    Control Plane relates the request to its owner, intended result, and scope.

  2. A coding agent prepares the change.

    The agent uses its own harness. Habitat associates execution with an Operation and its permitted resources.

  3. Other workers test and investigate.

    One runs tests; another examines an infrastructure dependency. Their work and handoffs stay related to the original request.

  4. A human reviews the privileged step.

    Production access requires a separate decision. Approval concerns the specific action and scope, not unrestricted credentials for the agent.

  5. An authorized worker performs the permitted action.

    Habitat checks runtime authority. Evidence connects observations to the worker, grant, and work; an unconfirmed external outcome stays unconfirmed.

Control Plane coordinates decisions. Habitat manages execution. Evidence supports checking the account. This example describes the intended product relationship, not universal integrations or production-wide enforcement.

04 / Authority follows the work

Delegation
has a scope.

The organizational decision describes the work being authorized and its limits. Habitat’s runtime binds bounded grants to the Operation and checks the current worker’s custody; delegation must not silently widen that authority.

The Control Plane coordinates shared policy and approvals. Habitat handles execution and runtime custody checks. External systems still enforce their own credentials and policies; a change in custody is not instant revocation everywhere.

Full credential brokerage and the broader connected approval workflow remain in development. This architecture is not a claim of production-wide enforcement.

Read the authority boundaries ↗
05 / For platform and security teams

Shared context.
Clear responsibility.

Platform teams get a common way to describe ownership, dependencies, and handoffs across agent tools and environments. Security teams can examine the scope of delegated work and the evidence supporting sensitive actions.

Habitat manages how the authorized work executes. Evidence makes the supporting account checkable. Control Plane connects those responsibilities to organizational decisions; it is more than a queue for launching agents.

Where does shared work
lose its context?

Bring a workflow involving several agents, a handoff, or a decision that needs human approval.

Talk about Control Plane ↗