The AI Agent Tutorial That Should've Been Your First (no code)
About this video
A beginner tutorial for building a first AI agent in n8n in three steps: give it a brain, give it tools, and teach it when and how to use each tool. An agent needs a brain (a chat model plus memory), tools such as Slack, Google Sheets, and Notion, and a ‘brain stem’ — the system prompt that relays instructions so the brain controls the tools correctly. Using n8n Cloud (14-day free trial; n8n can also run locally for free with technical setup), Jeff builds a subscription tracker: a chat trigger connects to an AI agent node, an OpenAI chat model (he picks a latest mini model, 4o mini, noting API usage is pay-as-you-go and separate from a ChatGPT Plus or Pro account), and Simple Memory with a context window he defaults to 14. A Google Sheet named ‘Subscription Tracker’ with columns Expense, Charge Date, Cadence, Cost, and Status is connected as a tool set to append a row, renamed ‘Add Entry’, with column values defined automatically by the AI. Rather than hand-writing the system prompt, he screenshots the workflow, downloads it as JSON, and asks ChatGPT’s latest reasoning model to generate the system message. The first test fails instructively: the agent adds an iCloud entry without confirming and uses a wrong date from 2023. He fixes the prompt to always send a confirmation message even when information is complete, and drags in a dynamic timestamp variable so the charge date reflects when the message was sent; the retest confirms first, then writes the correct row. He closes by sketching duplicate handling — add a get-rows tool, change the operation to append-or-update, and update the system prompt to check existing entries first — and warns that pre-built agents are black boxes that might hallucinate or leak private data.
Key takeaways
- An AI agent needs three parts: a brain (chat model + memory), tools, and a system prompt acting as the brain stem that controls tool use
- Memory matters: without it the agent forgot a list it had just acknowledged; a Simple Memory context window of 14 retained it across the conversation
- Generate the system prompt by giving ChatGPT a screenshot of the workflow plus its downloaded JSON, then paste the result into the agent’s system message
- Always require a confirmation step and use a dynamic timestamp variable for dates — the first test added an entry unconfirmed with a wrong 2023 date
- To prevent duplicates, add a read/get-rows tool, switch to append-or-update, and update the system prompt to check existing entries before writing
Episode details: published August 12, 2025 · 16:58 long. Summary written by Self Growth Videos from the episode itself. Business results and any income figures mentioned are Jeff Su’s own claims, shared for education — they are not typical or guaranteed.
← Previous: My Brutally Honest Advice About Joining Google (after 9 years) · Next: 3 AI Habits So Powerful It Feels Like Cheating →
All Jeff Su videos · Jeff Su profile · Jeff Su (official site)
