Let's start with webhooks. That's when one service tells another "X just happened, deal with it." Without webhooks (the old way), a script would hit the API every 30 seconds asking "anything new?" With webhooks it's the other way around — an event fires and the service itself pings the address you gave it / runs the command.
What you can build with webhooks (and what I've already tested myself):
- A payment comes in via [Stripe](https://docs.stripe.com/webhooks), the customer automatically appears in a Notion client database. No manual copying.
- Someone fills out a form on a landing page [in Tally](https://tally.so/), the team gets a Slack notification, and a task gets created in the CRM. (On my site [aigency.works](aigency.works) that's exactly how it works — form submissions hit my inbox and drop into the Notion CRM simultaneously.)
- A competitor posts a new video, the webhook fires, a Telegram bot pings you with a link.
Webhooks have been standard at GitHub, Stripe, Slack, Cloudflare, Vercel for ages. Now Notion has them too.
Next, Workers. A Worker is a piece of code that doesn't live on your laptop — it lives in a cloud provider. Cloudflare Workers, Vercel Functions, AWS Lambda all work the same way. You write a function, deploy it, and it runs itself whenever it's called. No server to maintain, no paying for idle time.
What you can build with a Worker:
- A price scraper. The Worker checks a competitor's site every hour and writes prices to Google Sheets.
- An AI wrapper. Your Telegram bot hits OpenAI through its own Worker. Easy place to bake in a system prompt, cache responses, and track token usage.
- A mini-API for a landing page. The "leave your email" form calls a Worker that sends a welcome email and drops the contact into Notion.
— On my personal site, a Worker handles serving OG descriptions in the right language to search bots. It also helps robots/bots/agents — depending on type — get exactly what they need (sitemap, llms.txt, etc.).
Third, CLI — command-line interface. That's controlling a program through text commands in the terminal instead of clicking buttons. The big advantage: terminals are something AI agents understand natively.
How I use CLI:
- I tell Claude Code to deploy the site, and it runs `vercel deploy` in the terminal itself.
- I ask it to "create a new GitHub repo," and the agent runs `gh repo create my-project` and pushes the code.
Without CLI, an agent would have to click around on screen. That's slower, breaks more often, and burns tokens. Though Claude Code does have a habit of reaching for browser use (it loves opening Chrome and taking screenshots). I've added instructions telling it not to do that unless there's a real reason.
Notion has its own CLI: `ntn.` Now Claude Code can deploy tools into Notion the same way it used to deploy sites to Vercel. (Vercel, by the way, says 30%+ of new deploys are already made by agents, not humans.)
Here's how I'm actually using this:
I write post ideas into a Notion database. A webhook catches the "new entry" trigger, a Worker takes the idea, runs it through an LLM with the right prompt, and writes the result to a "Drafts" database. By morning I've got three ready-to-edit post drafts waiting.
A Custom Agent in Notion (a configurable AI bot inside your workspace) with a Worker tool that hits CRM, Google Analytics, Cloudflare Analytics, and Stripe. Ask "which deals are at risk" and the agent pulls its tool, goes to the CRM, and returns real numbers — not guesses based on general context.
My landing page is built on Lovable, and the "submit a request" form calls a Worker. The Worker creates a card in a Notion database, sends a notification to Slack/Telegram, and fires a welcome email to the client. All without me touching anything.
Notion built all of this in partnership with Vercel. But right there in the README they write: "you probably shouldn't use it for anything serious just yet." Still being baked, basically.
Who this is useful for right now — IMO, people already building automations in Claude Code via terminal and wanting them to live inside Notion. Anyone whose business runs half on Notion and has a free evening to experiment. Everyone else should probably wait for the general public release.