If I custom build an AI agent using code, how can I give it to the client?
Well, ideally you would build the AI agent using code directly in the client’s environment, so you’d create the NADN AI agent there or use the NADN code block inside their environment. That way you never have to hand over anything because it already lives in their system. This is one reason I recommend getting the client to pay for the services and handling everything during a kickoff video call where you guide them through the process. You avoid a handoff, which can be annoying. It adds a bit more friction up front, but it saves a lot of hassle later and lets you use things like affiliate codes. I would completely eliminate the handoff.
He recommends a two‑step process: first, when you get the money, book a kickoff call; on that call, have the customer sign up directly to all the platforms you’ll use, capture their username and password, then proceed with the standard SOP (which is the same regardless of automation platform). He likes this because it’s simple and lets you also get the customer to sign up using your affiliate code. This eliminates the handoff problem—clients front‑load the work, you earn affiliate revenue, they retain full intellectual property, and you avoid any handoff. When delivering, you simply record a video walking the client through the system, explaining how to access it and providing any necessary links.
Requiring clients to create their own Make (or NADN) account and granting you access is non‑negotiable. It simplifies handoffs, speeds up onboarding, and reduces liability because you avoid persistent recurring costs on your side. You can also add affiliate links to boost margins by 3‑5%. I’ve tried pass‑through billing, but managing everything on the client’s account is more transparent and avoids the dark‑pattern risk of owning their business infrastructure.
On the client's. I'd recommend looking into Claude Routines — a workflow offering Anthropic launched a couple months ago; it's essentially a skill that runs in the cloud on a loop (you set the loop interval, or trigger it via API). It costs more in tokens since you're running AI inference, but it's straightforward. Delivery works the same way as delivering an n8n or Make.com automation: build it in their workspace, then record a walkthrough video showing them the whole system end-to-end so they can run it independently with or without you — it's their IP, and making things stable long-term for the client is part of the job as a service provider. After the delivery video (also a great relationship-building moment), always try to upsell: point out something you noticed while building that could make them more money with a few tweaks, and offer a call to talk through it.
claude codeclaude routinesclient deliveryautomation ownership
If you're building a system that generates $5,000 a month in value for a client, having them pay a small percentage fee is not a big deal; they care about the upside, not the tiny downside. As the service provider, you might worry about subscription costs, but the AI and automation field attracts technical people who overthink this. Get the client to pay for everything. On a kickoff call, walk them through exactly which platforms they need to sign up for. This builds perceived value because they see the technical setup and appreciate you handling it. You can also use your affiliate links—being a Make.com partner, for example—to earn 3–5% kickback or give clients discounts. If a client stays for a year, you earn recurring affiliate fees; e.g., 40% on a $100/month plan yields $40/month, or $480–$500 per year per client even if they stop working with you. This approach also minimizes your liability since you’re not making monthly payments on their behalf. It’s the best way to handle software costs. If you’d like a kickoff call SOP, I have a document that outlines how to run the call and get clients signed up on the platforms.