PROMI orchestration

The operating loop

PROMI gives AI work a visible lifecycle: scope, plan, route, execute, verify, review, and continue. Cortex remembers the decisions and evidence; PROMI uses that memory to decide the next safe step.

• Scope the goal and project boundary.
• Break broad intent into bounded work packets.
• Route each packet to the right role or AI worker.
• Verify claims before treating work as complete.

Handoffs and ownership

A handoff records who owns a piece of work, what is in scope, what evidence is expected, and what should happen next. That keeps workers from colliding and gives reviewers a practical way to inspect progress.

• Every meaningful work packet should have a visible owner.
• Completion should include evidence, not just a status change.
• Residual risk should be recorded before the next reviewer acts.

How PROMI uses Cortex

PROMI coordinates the work, while Cortex remembers the context. PROMI should use project memory, handoffs, evidence, and prior decisions to decide what can move forward and what needs a human review gate.

What users should see

Users should see a clear project state: what is planned, what is active, what is waiting, what is blocked, what is ready for review, and what decision is needed next.

Progress recovery

When work stalls, PROMI should help recover the project by identifying the blocker, creating the next safe handoff, asking for clarification, or routing the work to the right role.

Heartbeat and automation

The dedicated per-project runtime adds a heartbeat, scheduled tasks, and event-triggered handlers. That lets PROMI run recurring checks, respond to project events, and surface blocked or stale work without requiring a user to restart the process manually.

• Scheduled checks can cover compliance, tests, quality, stale handoffs, and memory freshness.
• Events such as commits, completed tests, milestones, and new handoffs can trigger follow-up handlers.
• Automation should remain visible: users should see what ran, why it ran, and what it found.

When PROMI pauses

PROMI keeps routine work moving, but it should pause when judgment is required. Ambiguous scope, blocked work, failed verification, customer-visible changes, deployment, billing, authentication, data, or destructive operations should become review gates instead of hidden automation.

Review-ready evidence

When PROMI says work is ready, the review package should answer what changed, why it changed, how it was verified, and what risk remains. If that evidence is missing, the work should not be treated as complete.

What users can configure

Users and customer admins should be able to configure project owners, review gates, notification expectations, support/escalation paths, and which kinds of changes require approval before continuing.

What can go wrong

PROMI can only coordinate well when goals, owners, and review expectations are clear. If work feels stuck, check whether the blocker is missing input, failed verification, unclear scope, or a decision that needs a human owner.

Read next

Read Harness engineering next to understand how Kaidera keeps work bounded while AI workers execute tasks.