AI Project Delivery and Handover Playbook
Full lifecycle framework for shipping client AI workflows without headaches. Key rules: (1) Clients host their own n8n instance (cleanest licence compliance and security story), (2) Clients own and pay for all API keys, send a Loom walking them through key creation, never become the billing babysitter, (3) Run internal QA for several days with logged inputs and outputs before client QA, (4) Use test vs production environments and back up workflow JSON to GitHub or Drive, (5) Define done as a bulleted functionality list in the scope of work, (6) Feature requests mid-build go on a v2 backlog, not in the current scope, (7) Final deliverables: clean sticky-note documentation, Loom walkthrough, maintenance retainer option (typically 10 to 25 percent of original project fee per month), and a written exit process for IP and off-boarding.
The 10-hour Claude Code course repeats this delivery pattern in the context of agentic workflow projects: decide where the workflow will live, secure data and credentials, test with realistic client data, log inputs/outputs/tool calls/errors, separate test and production versions, document the system, define ownership/IP/exit terms, and keep maintenance retainers separate from new feature work [src-016].