Pages are becoming teammates

Happy Tuesday.

New episode dropped. I sat down with Josh Puckett. He's designed at Dropbox and Wealthfront, placed 100+ designers through Upper Study, invested early in Framer and Runway, and just shipped Interface Craft.

We talked about what high slope actually looks like, how to practice uncommon care, and whether "AI native" is a real skill or just a moment we're passing through.

One line that stuck: if you think your job is getting paid to create an interface artifact, you're going to have a very bad time. The cost of that artifact is now zero.

Which is exactly the thread running through what Notion just shipped.

– Tommy (@designertom)

TOGETHER WITH NOTION

This issue is brought to you by Notion’s Custom Agents.

Notion's new Custom Agents run recurring work across Notion, Slack, Mail, Calendar, Figma, Linear, and MCP-connected tools.

Use them with scoped permissions, logged runs, and usage controls.

If your team is moving from one-off AI prompts to shared operational workflows, this is one of the more significant launches this year.

Try Notion Agents →

The Page Becomes a Teammate

Most design teams use their workspace the same way.

The brief lives in a doc. Research lives in another doc. Decisions, if they get written down at all, live in a third. The backlog sits in a database somebody set up once and nobody maintains.

AI showed up and made some of that faster, but usually for one person at a time.

You open chat, ask a question, generate a summary, clean up a draft. It’s useful, but still reactive and single-player.

I think Notion just solved that at scale, and we’re using it at our studio.

Custom Agents are letting our team build agents with their own instructions, triggers, tool access, and permissions, then share them inside the workspace like any other page.

They’ve framed these as autonomous teammates: agents that answer repeat questions, route tasks, compile updates, and work across Slack, Mail, Calendar, Figma, Linear, and custom MCP servers.

Internally, they already have more Custom Agents than employees. And early testers have built more than 21,000 agents already.

A huge chunk of design work is coordination work. As designers, we feel this more than anyone:

• Finding the latest context

• Routing feedback to the right person

• Turning a Slack thread into a ticket that actually gets triaged

Making a decision visible so multiple people can move without stepping on each other. If you work on a team, you've probably been the human router on your team at some point.

In my conversation with Notion's co-founder, he told me "permissions behave like pages."

One agent stays personal. Another gets shared with the design team. A third runs in the background for a specific Slack channel or project database.

This is how we work with our research process, too.

Each one is scoped the way you'd scope a doc (who can see it, who can edit it, what it has access to). That is a different model — in a real way — from the all-powerful company bot most teams are quietly uneasy about.

I’m seeing some interesting workflows emerge:

1. A design-research agent that watches new interview notes and tags patterns into the right project space.

2. A design-ops agent that turns stray Slack feedback into triaged backlog items.

3. A launch agent that pulls weekly status from GitHub, Figma, Notion, and calendar context so nobody has to be the person who writes the update by hand.

4. A team knowledge agent that handles "where's the latest file?" and "who owns the component?" without someone volunteering as the human index every week.

This is the “toil” that typically eats human hours away from the interesting work.

The question for design teams is: what happens next?

Once the agent can draft, summarize, sort, and route, the leverage shifts from execution to orchestration. Can the team see what the system did? Can they trust it? Can they shape it without every workflow turning into invisible background magic?

That's where designers have the most to gain — and the most to contribute.

If agents run in the background, people need a clear mental model of triggers, scope, cost, edit rights, and failure states. They need to know who changed the agent, what sources it touched, and when it last ran.

The page becoming a teammate only works if the teammate is legible.

For years, software teams used documents as a source of truth. Then the doc became a workspace. Now the workspace is starting to take action on its own.

The designers who matter most in that world will be the ones who define scope, structure permissions, map the handoffs, and make the machine behavior understandable to everyone else on the team.

That's it for this one.

When the workspace starts acting on its own, the designer's job grows. More responsibility over the system around the work — what an agent can see, what it can do, and whether anyone on the team trusts it enough to let it run.

If you've already wired an agent into a design workflow, reply and tell me the first repetitive job you handed off.

See you soon.

Tommy