高見龍：當我把事情交給 n8n 之後，人生變得更廢了

1. In one sentence

n8n is a low-code tool that breaks repetitive work into individual nodes and chains them together to run automatically. The point is not AI, it is the ability to decompose a task — and you can pick it up with no engineering background as long as you are willing to start with something small from daily life.

2. The session

The host introduced him as a repeat champion of iThome's marathon contest and the author of bestselling technical books including *Learn Git for Yourself*, *Learn Ruby on Rails for Yourself* and *Learn Python for Yourself*, currently a co-founder of 五倍學院 (on verification, the official education brand page uses the name 五倍學院 for the organisation; its team page also lists him as its principal, so the two titles may correspond to founder and current role — noted here as well).

What n8n is, and what it is not

The name n8n is "node" and "automation" mashed together (the intended coinage was "nodemation", but with an n at each end and eight letters in between they could not be bothered typing it and shortened it to n8n). He defines n8n as a no-code, low-code tool where "a pile of nodes combine to complete a task", and made a point of clarifying: n8n is not an AI tool, it just happens to have an AI node among its many nodes; its real value is task decomposition.

His example of what decomposition means: suppose your boss asks you to "check PTT and Dcard every day for anyone slagging off the company". That splits into four nodes — "every day" is a timer, "go to PTT or Dcard" is a web scraper, "is anyone slagging off the company" needs AI for semantic analysis, and "report to the boss" is an email notification. He also said bluntly that n8n's interface is engineer-oriented and there are four to five hundred nodes, so it is natural to feel lost at first — but his advice is simple: just try it, and the worst that happens is you break it and reinstall.

A live demonstration: scraping the conference site's speaker list

He demonstrated a small workflow live: call the conference site to get the page HTML, parse the speaker names out with a selector (he admitted his selector was fragile and would break the moment the site was redesigned), then post the full speaker list to a Discord channel, using the array's length to count the total.

Solving his own problems

He shared several small annoyances he uses n8n for day to day:

• Book ranking monitoring: he habitually checks where his "learn for yourself" series ranks at the Tenlong bookshop each day, so he built a workflow — a daily timer, scrape the site, match keywords, send a Discord notification. It notifies whether or not the book charts, because if he ever stopped getting a notification he would not be able to tell whether it was "not charting" or "the flow itself broke".
• Backing up workflows and keys: since n8n is an automation tool, he uses n8n to back up n8n's own workflows and keys on a schedule, running a system command to export them and sending a Discord notification, so he is not stranded if the system ever fails.
• Client message board to Excel: a client wanted the message board data exported to Excel so they could email thank-yous to the people who posted, which he did with a database connector node plus a file conversion node; the key part was a webhook node with basic auth for permissions, handing the client the link directly so they could enter a username and password and download it themselves. The whole workflow took under ten minutes and needed no deployment.
• Translating a manuscript into English: he is writing a book for an overseas publisher, with a Markdown manuscript containing many localised examples, code blocks whose formatting has to be preserved, translation that must not embellish, and a tone that has to stay consistent across chapters — and different large language models translate to different standards. He built a form node to upload the Markdown file, wired AI processing in the middle, and gets the translated Markdown back; he spot-checked a few chapters, confirmed the quality was stable enough, and adopted it directly.

Solving clients' problems: media, schools, hospitals

• Media: journalists need to know immediately whether specific keywords (the India-Pakistan conflict, a particular person's remarks) or specific institutions (the Presidential Office) have published anything new. Because most of these sites and institutions offer RSS, the approach is relatively simple: a timer plus an RSS node to fetch sources, keyword matching, and routing the notification to different Slack channels by keyword.
• Schools: a library wanted an online customer service bot answering FAQs. He pointed out that for a setting with little data and simple content you do not need RAG or a vector database at all — an HTTP request node to scrape the FAQ page and an AI agent node answering from that content is enough.
• Hospital pharmacists: a pharmacist wanted to check every day or two whether the Food and Drug Administration site had announced new drugs or withdrawals, but the site offers no API, so the page parsing logic had to be written by hand; he also used a spreadsheet as a rudimentary database, comparing each fetch against what was stored previously and only sending a notification when something differed — the most effortful step in the whole flow.
• Hospital nurses: nurses have to tick and notify high-risk patients meeting particular conditions (blood pressure, heart rate values and so on) from a spreadsheet of patient data every day. He built a form node for them to upload the spreadsheet, with conditional and code nodes filtering on values in the middle and a file export at the end, shrinking work that used to take a while by hand into a few seconds.
• A student's request (declined): a student asked whether n8n could help snap up concert tickets, and he refused to teach it on the spot, citing Article 10-1, paragraph 3 of the Cultural and Creative Industries Development Act directly: using false information or other improper means to buy performance tickets or obtain booking or collection credentials via a computer or related equipment is punishable by up to three years' imprisonment, or a fine of up to NT$3 million, or both (on verification, the text on the Laws & Regulations Database has paragraph 3 reading exactly that, matching what he said on stage). He stressed that he later took a different approach: only monitoring the remaining ticket count shown on the sales page and sending a notification when it changes, rather than automatically buying.

How to get started with n8n

He suggests running one command on your own machine (npx n8n), which gives you a working local n8n environment within minutes; the drawback is that closing the laptop stops the workflows, so running around the clock means either never shutting down or moving to a cloud host. He also cautioned against trying to save the world with n8n on day one, recommending starting with the smallest chores in your own life to build a feel for it, and taking on bigger problems once you have accumulated some sense of achievement.

Frequently asked questions

Asked "can I learn this without a technical background", he said he could not promise, but learning a bit of Python for self-defence is no bad thing in this era, and gave the URL of his own free, open-source introductory material. As for "will AI agents replace automation tools like n8n", he sees them as different in kind: AI carries a degree of non-determinism, whereas an automation flow like a backup can have none at all — one failure and you are in trouble. He positions n8n as the "glue" between applications, the point being to learn how to connect different services rather than to replace AI or be replaced by it.

3. Figures and cases

• Global search volume for n8n rose noticeably from the start of that year, an observation he drew by comparing Google Trends against Zapier (speaker's own account, no external source).
• Roughly four to five hundred node types.
• The client message board to Excel case: from idea to finished workflow in under ten minutes, per the speaker (speaker's own account, no external source).
• The nurse patient data case: manual work that used to take about half an hour, shortened to a few seconds with the workflow (speaker's own account, no external source).

4. Lines worth keeping

• "You don't have to be great to start, but you have to start to be great."
• "n8n isn't AI; it just happens to have an AI node built in."
• Quoting *Frieren: Beyond Journey's End*: "Magic is a world of imagination; anything you cannot imagine in the world of magic cannot be realised."

5. Tools and terms mentioned

n8n, node, workflow, timer node, web scraper node, HTTP request node, RSS node, webhook node, basic auth, AI agent node, if node, code node, npx, Node.js, Zapier, Make (formerly Integromat), PTT, Dcard, Discord, Slack, RAG, vector databases.

6. Wider observations

The value of this session is not the tool but the mindset of task decomposition: turning a vague request from your boss into four nodes — timer, scraper, semantic judgement, notification. That decomposition ability is the same logical training as programming, only without actually writing code. The way he distinguishes n8n from AI agents (an automation flow cannot tolerate non-determinism, AI can) is almost exactly the logic Wisely Chen used earlier about the difference between RPA and AI agents, and set side by side the two corroborate each other. His refusal to teach the student ticket-snapping was handled gracefully too: state the legal risk first, then demonstrate an alternative that monitors and notifies without automating a purchase — a demonstration that "the client's responsibility" is not only accepting the result but judging what should never be automated at all.

7. Sources