What happens to my AI when the consultant leaves
If the implementation was done correctly, nothing breaks. The brain lives in the firm's own folder, on the firm's own stack, Microsoft 365, Google, iCloud, and the prompts, workflows, and operator instructions are plain files the team can read. If Quiet Machines disappeared tomorrow, every implemented firm would still have a working AI brain on Friday.
Why it survives the consultant leaving
The implementation is folder-native. There is no Quiet Machines server, no proprietary system to log into, no portal that goes dark when a subscription lapses. The firm's prompts, knowledge base, voice training, and operator playbooks are markdown files inside a folder the firm already owns. The AI model itself, Claude, is purchased on the firm's own Anthropic account, not on the consultant's account.
The handoff that proves it
Every implementation ends with a printed Operator Runbook, a recorded training loom, and an "If You Fired Me Tomorrow" one-pager. That last document literally lists every credential, every file path, and every workflow with the steps to run it without a consultant present. It's the closing exhibit of the engagement.
What's actually in the folder
- The Client Brain. Markdown files describing every active client, household, goals, last-touch, planning state.
- The voice library. Every quarterly letter, blog, and email the firm has ever published, indexed for the model to reference.
- The operator playbooks. Step-by-step prompts for meeting prep, lead scoring, content drafting, compliance review.
- The compliance guardrails. What the brain will not say, what it always must say, who reviews what.
What "if you fired me tomorrow" actually covers
The handoff one-pager is one printed page that lists: where every file lives, which Anthropic account holds the model subscription, who has admin on the folder, and the three commands a non-technical team member can type to keep the brain running this week. It's intentionally short. If it took ten pages to hand off, the implementation was built wrong.
What can break
The model itself can change, Anthropic ships new versions of Claude every few months. Prompts sometimes need a small tune-up. That's a half-hour job for anyone with the runbook in front of them. Nothing structural breaks.
Why the implementation is built this way on purpose
Most AI-for-advisor products are subscription portals, the firm pays monthly for access to a vendor's cloud, and if the firm stops paying or the vendor pivots, the value evaporates. The QM implementation is the inverse on purpose: nothing about it depends on QM being around. The firm pays Anthropic directly for the model, owns the storage, and runs the brain on its own credentials. That structure is partly a compliance stance (SEC books-and-records favors data the firm can produce on demand) and partly a trust stance, a consultant who can walk away and have the work survive is a different kind of consultant than one who can't.
What an implemented firm should actually do month one without the consultant
Run the brain on a full week of normal work without calling QM. Find three things that are wrong or unclear. Open the runbook, follow the troubleshooting page, and fix them. That exercise is the test that the handoff worked. Every QM-implemented firm is asked to run it within 30 days of the on-site, while the implementation is fresh and Lights-On is still active, so any gaps surface while support is still standing by, not six months later when the firm is alone.
Sources
An implementation you can't leave with isn't an implementation, it's a subscription. AI visibility audit →