Introducing Plan Mode. Because a prompt without a plan is just a wish. In plan mode, your agent shows its work in a read-only sandbox before touching anything. Just plan → approve → execute.

Had this with Claude Code for a while. The moment an agent shows you its plan before touching your code... you never go back. Glad to see Notion shipping it too.

W Notion 🤍🖤!!

The real ones saw this in app days ago 🫡

Awesome

🫡

why i’ve deleted my notion account. yall pivoted from what made yall great to desperately wanting to be like openAi or Anthropic. we dont need these agents to be shoved down our throats. an MCP would’ve sufficed.

Notion Playground in full swing

plan before execute is the right paradigm. especially in high-stakes workflows, agents need a review layer before touching anything. this is how trust gets built gradually. legal work is the ultimate test case for exactly this pattern at @GenieAI

The trust win here is not just planning. It is making the no-edit state explicit. Teams adopt agent workflows faster when review and approval boundaries are obvious before the first change, not after a surprise diff.

Adding a planning step can introduce friction. For simple tasks, it may slow things down more than it helps, so the real value shows up in complex, high-stakes workflows.

You guys are giving FedEx a run for its money - shipping this much everyday.

Up until now I've been saying "I've got this idea about X, ask me clarifying questions to help me flesh it out." Looking forward to using this. A lot.

Super cool. Question: How is it different from just asking it to write a plan on a new page before touching anything?

This is exactly what we need for automation tools. I've built AI workflows that would randomly change things without warning - terrifying when you have real business operations running. The sandbox preview solves the "did my agent just break everything?" anxiety.

A prompt without a plan is just a wish" Spot on. This "Plan -> Approve -> Execute" flow is exactly the "Human-in-the-loop" architecture we need for safe Agentic AI. I was just discussing this shift today: we are officially transitioning from "Prompt Engineers" to "Workflow Reviewers." Brilliant update to see this natively integrated into Notion's sandbox!

This is what I have been looking for. I want to plan my schedule throughout the weeks and the flexibility to change each task.

For real though… all these updates lately are really making a strong case for Notion. Glad I finally updated my payment method… 😅😏

make a cli

Sandboxed execution is the right foundation either way. Plan-then-approve for high-stakes decisions. Autonomous scheduled runs for recurring work. The pattern isn't one or the other. Best agents do both depending on the task.

Read-only previews stop agents from breaking prod. Treat it like a dry run, not a UI toggle.

copy paste plan mode from claude to notion. Claude went worst in the last 1-2 weeks, latest with Opus 4.7. When will notion follow? 🙈💣

That sandbox planning is pure gold.

Allow Notion to connect and orchestrate workflows across apps running on the local machine. Im sure this will help

Add an auto review mode after plan execution too. Too often AI execute 80% of the plan and miss the rest subtly

Love the sandbox before execution—how's approval work for multi-step plans?

@NotionHQ no ChatGPT 5.5 on launch day ?

"A prompt without a plan is just a wish." That line alone is worth bookmarking. Running 23 agents, the ones that work aren't the smartest. They're the ones with the clearest instructions written before anyone touched a keyboard.

exactly. once the agent can touch state, the approval step stops being a UX detail and becomes the product.