Skip to content

Capability planning

A plan is only useful if it can be executed safely and checked afterward. The ecosystem separates five steps that are often collapsed into one prompt.

1. Goal

What outcome is wanted, under which constraints? A goal is not a plan. It should state the desired result, the boundaries, and the evidence that would count as success.

2. Plan

A plan turns goals, state, capabilities, and constraints into a structured, revisable artifact. It should be explainable and verifiable before execution.

Project: ORCHADYN.

3. Capability governance

Before a step runs, the system needs to know whether the capability exists, who may invoke it, under which policy, and with what audit. A capability catalog makes this explicit.

Project: MODUREGIS. Authorization is evaluated by AEGIVELA.

4. Execution

Long-running work needs durability: retries, compensation, recovery, and a recorded history. A process instance should survive a crash without repeating side effects.

Project: RHEOVELA.

5. Human decision and verification

High-risk steps may need approval, exception handling, or intervention. The outcome may need verification against the original intent.

Projects: SYMBIVELA for human control, TEKMOVELA for verification.

Why separate the steps

  • Each step has a different artifact. Goal, plan, capability decision, process history, approval, and verification result are not the same thing.
  • Each step has a different owner. Business, architecture, platform, security, operations, and assurance teams can own their own boundary.
  • Each step can be inspected. Collapsing them into one model call makes the decision opaque and the evidence weak.

A practical test

For a proposed autonomous process, ask:

  • Can I see the plan before it runs?
  • Can I name the capability and policy for each step?
  • If step 3 fails, can the process recover without repeating step 2?
  • Who approves the high-risk step?
  • What evidence proves the outcome?

If the answers are “it is all in the agent,” the process is not yet plannable, governable, or verifiable.