The Front Door and the Workshop

This morning I opened up Claude, started a voice conversation, and told it I wanted to keep working on a draft for my blog on Buzzy, the blogging tool I built.

It automatically reached for the Buzzy connector, read my writing guide, and pulled up the draft. Then we talked through the post in voice mode.

I never opened my own app until I reviewed the draft.

And that’s when something clicked for me; it's imperative to design for experiences that start from a user's AI interface of choice.

The airline problem

Remember booking flights directly on airline websites? The airline owned the experience. The upsells, the loyalty program, your attention.

Then comparison sites showed up. Airlines still flew the planes, but someone else owned the front door. And with it, a big part of the customer relationship.

Agents are doing something similar to software.

Instead of opening Salesforce or HubSpot, you tell an agent to follow up with leads, chase unpaid invoices, and update the CRM. The software still does the work. You just never see it.

That’s a pretty big problem for companies that spent decades optimizing their interfaces, onboarding flows, and per-seat pricing.

If one agent can do the clicking for an entire team, what exactly are those seats worth?

The software doesn’t become worthless. But the value shifts. The company that owns the interaction increasingly controls how the underlying software gets used.

So what happens to the app?

I think it becomes something more interesting.

The last 30%

Here’s how I actually work.

At Magicare and with the startups I design for, agents can take a design task surprisingly far. Give one a good prompt and it gets me maybe 70% of the way there. Add a design system and it gets even further.

Then I take it back to Figma, Cursor, or Claude and start pushing.

The remaining work isn’t just polish. It’s where I question the assumptions, explore alternatives, and figure out whether we’re building the right thing in the first place.

An agent can generate a hundred ideas. Experience and taste help me recognize which one is worth pursuing.

I recently worked on a mobile app redesign where the agent handled a lot of the tedious work. Auto layout, components, all the little things that used to eat up days.

That gave me more time to think about the product itself.

I wanted the agent to be accessible from anywhere in the app, without forcing people to leave what they were doing and open a separate chat.

So I designed what I call a “trunk drawer.”

Most apps have a drawer you pull down from the top to access filters or additional information. This one works from the bottom. Swipe up from anywhere and the agent is there.

It knows which screen you’re on and what you’re working on. It can answer questions, kick off workflows, and show what’s happening in the background, a little like Apple’s Live Activities.

The point wasn’t to add another AI chat window. It was to make the agent part of the application itself, always available without getting in the way.

Could an agent have proposed something similar? Maybe. But deciding that this was the right interaction, and how it should work within the rest of the product, was the design challenge.

And that’s what I think the future of the app looks like.

The agent is the front door. The app is the workshop.

The workshop isn’t going away

When agents can handle most routine tasks, people won’t need to open applications nearly as often.

But when they do, they’ll want something different.

They’ll want to explore, make decisions, inspect what’s happening, correct mistakes, and take control.

Think about a designer working in Figma, a developer in Cursor, or someone reviewing a complicated financial model. These aren’t experiences you necessarily want to reduce to a conversation. Sometimes you need to see the whole thing, move things around, compare alternatives, and work directly with the material.

That’s the workshop.

And I think the better agents get, the more valuable a well-designed workshop becomes.

Not because humans need to do every task themselves, but because delegating more work makes judgment and control more important.

The workshop needs depth, flexibility, and precision. It also needs agents built into it, ready to take over the tedious parts without interrupting the human parts.

That’s what I was trying to solve with the trunk drawer.

The agent doesn’t replace the interface. It moves between the front door and the workshop, carrying context with it.

Design both doors

Here’s the part I think designers need to pay attention to.

For a lot of people, the first interaction with your product won’t happen on a screen you designed.

It’ll happen through an agent.

That means your product’s first impression might be a tool call. Your onboarding might be a set of instructions an agent reads. Your error messages might be written for software that needs to understand what went wrong and what to do next.

When Claude connected to Buzzy, it received a writing guide before touching my draft.

That’s onboarding. Just not for a human.

And it needed design decisions: what context to provide, what actions to expose, what defaults to set, and when to hand control back to the user.

We’ve spent years designing interfaces that make software easier for people to use. Now we need to design interfaces that make software easier for agents to use on our behalf.

But these aren’t two separate products.

They’re two ways into the same product, and the transition between them matters just as much as either experience.

Front doorBoth the same product and contextWorkshop

• Front door: agents doing routine work
• Workshop: humans taking control
• Both: the same product and context

I started editing my blog through Claude. I finished reviewing it in Buzzy. The context and the work moved between them.

That’s the future I see. Agents handling the routine work through the front door, humans stepping into the workshop when they want to think, create, or take control.

Apps aren’t disappearing. Their purpose is changing.

Most designers have spent their careers designing one door. Now there are two, and the one most people walk through first was never on our Figma canvas.