I'm working on a 3k AI chatbot system...some things you need to know
A few days ago I landed a client to help them set up an AI agent for email, Facebook, and Instagram. Working on this project taught me lessons most YouTubers don't share. This is knowledge from the trenches. Here we go. When building an AI agent, combine the system prompt and the user prompt in the same field in n8n. This gives the AI a better response to the customer. 2. Spread the information across multiple agents. Create a billing agent, a tech agent, and so on. This stops you from stuffing one agent with too much information, which leads to misfires. 3.Always use Sonnet 4.6 or higher for customer service work. It is the best model on the market for this type of work right now. 4.When updating your prompts, focus on the start and end. Add all rules and constraints there. Sometimes repeat the same rule at both ends. The AI tends to glaze over the middle section. Use the middle for FAQ information and knowledge. The AI will draw on that to write its own responses. 5.When building your knowledge base, do not write it as Question and Answer format. Do not write it like: "When a customer says hello, answer with hello back." Write it so the AI can use the data to form its own answer from the available knowledge. For example, I write something like this: - **Topic: Ball Slicing (Driver)** - **Terms**: "stop slicing with driver", "driver slice", "curving right" - **User Intent**: Ball curves excessively to the right (for righties). - **Advice**: Strengthen grip slightly, aim a bit left, and shallow the swing path. Shorten the backswing and prioritise tempo over raw speed. When you write it this way, the AI reads the user intent and writes a message based on it. It is not rigid. I hope this helps someone when they start building their own agent!