Autonomous Orchestration

How AIPM keeps your projects moving on its own — doing the work that isn’t waiting on you, and notifying you (never deciding for you) the moment something needs your approval.

Status: in development. The design of this feature is agreed and it is being built now. The behavior described below may not be fully live yet — treat this page as a guide to what autonomous orchestration will do, not a promise that every part is switched on today.

What it does

Autonomous orchestration lets your project assistant keep things moving without you having to nudge each step. On a regular tick, the project-manager assistant looks at what work is ready, then either does the project-management work itself or launches a helper to carry it out — whenever that work is not waiting on you.

The point is steady forward progress between your visits. You stay involved only where your authorization or a genuine decision is required; everything that can safely proceed on its own, does.

Two assistants, one project

AIPM can run two assistants at the same time — a Claude assistant and a Codex assistant. When both are running on the same project, they sort out their roles automatically:

They coordinate with each other in the background, so both can work the same project at once. The result is simply more throughput — more gets done in parallel — with no extra effort from you, and with one assistant always clearly in the lead so there is no confusion about who is in charge.

It never approves on your behalf

This is the most important thing to understand: the system never approves, signs off, or makes a decision for you. Anything that needs your approval, your sign-off, or your input still waits for you.

When autonomous orchestration hits something that requires you, it does exactly one thing: it notifies you. It does not change an approval, advance a gate, or sign anything. Only you — the human sponsor — can approve. An assistant agreeing that something looks right is just agreement; it is never permission to proceed.

The guardrail, in one line. Autonomous orchestration can do work and can tell you when you’re needed — it can never grant your approval. Approval is yours alone, every time. (See also Approving with Evidence.)

How you’re notified

When something needs your approval or input, you’re notified in the local app first, and only escalated to your remote channels if you don’t respond there:

To act on it, you simply log in, review the pending item, and approve or reject it as usual — the same review-and-decide flow you already use. Each item notifies you once; you won’t be buried in duplicate alerts even when both assistants are running.

What this means for you, day to day

The system can…The system can never…
Do project-management work on ready, unblocked items. Approve, sign off, or decide anything on your behalf.
Launch helper agents to carry out work in parallel. Advance a gate or change an approval’s status by itself.
Notify you — in-app first, then push and email if you don’t respond — when you’re needed. Treat one assistant agreeing with another as your permission.

In short: it moves the project forward where it safely can, and it brings the things that need you straight to your attention — but the decisions stay yours.