Notion AI Office Hour Session With Nasri

Completed25 Jun 2026

Session recap

This week's Notion AI Office Hour ran live over Google Meet and was livestreamed on YouTube (Building with Nasri). True to form, it was kept raw and interactive — no fixed agenda, with the attendees steering where the hour went.

We opened with a quick introduction (five years in Notion, from aviation security supervisor and GrabFood rider to Notion consultant, ambassador, and co-founder of Workflowers), then ran two warm-up polls to set the tone — one on what people most wished AI could do for them, and one on where they'd place themselves on a simple AI maturity scale.

The four levels of AI usage

Level 1 — using AI just for chatting.
Level 2 — using AI for tasks and productivity.
Level 3 — running autonomous agents that work in the background.
Level 4 — a full business operating system powered by AI.

Over half the workforce sits at Levels 1–2, and only around 2% reach Level 4. Lenny was our resident Level 4 user, running AI across day-to-day work.

Two types of agents — and no model lock-in

We drew the line between conversational agents (you chat, they act — ChatGPT-style) and autonomous agents (they run quietly in the background while you sleep). A key point: Notion agents are LLM-agnostic, so there's no model lock-in — you can reach for GPT 5.5, Opus 4.8, Claude 3.1 Pro, and others as the task demands.

Notion as the memory layer for AI

Models like Claude don't hold persistent memory, so we framed Notion as the context and memory layer — structured databases the AI can read from and write back to, quickly. The practical lesson: don't hand the AI your entire workspace. Limiting its context keeps responses faster and cheaper.

Connecting Claude (and friends) via MCP

MCP is best thought of as a bridge or handshake — it gives an external AI the tools to read and update Notion. Connecting is simple (Settings → MCP → connect to Claude), and you don't need a premium plan for the connection itself. The catch we kept coming back to: build your database structure in Notion first. Claude can't reliably design Notion structure on its own, so it works best updating a container you've already shaped — exactly the advice for Tia's journaling workflow.

Live build: databases, workers, and a self-updating watchlist

We built databases from scratch using plain natural language ("build me a task database"), then showed a favourite: a movie watchlist where typing a title auto-fills the details. That's not AI burning tokens — it's a worker (a small script) quietly pulling from a film API in the background.

Workers vs agents, and keeping costs sane

This led into the money question. Workers run as code and cost a fraction of an agent because they consume no AI tokens. Agents run on credits (300/month on the business plan, then $10 per 1,000). A Canva image task, for example, ran at 34 credits — roughly 34 cents. The other lever is model selection: heavy models (Opus 4.8) for creative and reasoning work, lighter ones (Sonnet, Haiku, Gemini Flash) for simple steps like formatting or adding slugs. And on the free plan, you get roughly 2–3 hours of AI before the trial locks — handy to know when experimenting.