The PM’s New Design Loop: What Claude Design Changes About Shipping

For most of the last decade, the bottleneck between a product idea and something you could show someone wasn’t thinking. It was production. A PM had the insight, the designer had the bandwidth, and the gap between those two realities was measured in Jira tickets and Slack pings.

Claude Design collapses that gap.

Anthropic Labs launched it on April 17, and the short version is this: you describe what you want, Claude builds a first version, and you refine through conversation, inline comments, direct edits, or custom sliders Claude builds on the fly. It reads your codebase and design files during setup, so every project inherits your brand automatically. When a design is ready to build, it packages into a handoff bundle you can pass to Claude Code with one instruction.

That’s the feature set. The interesting part is what it does to the product development lifecycle.

The shift: design leaves the designer’s queue

The old loop looked like this. PM writes a spec. Designer mocks it. Review cycle. Engineer estimates. More review. A prototype shows up two weeks later, the feedback comes in, and half of it reveals the spec was wrong to begin with. Rinse, repeat.

Datadog’s PM Aneesh Kethini put the new loop bluntly in Anthropic’s launch post: what used to take a week of back-and-forth between briefs, mockups, and review rounds now happens in a single conversation. Brilliant’s senior product designer Olivia Xu said their most complex interactive prototypes, which previously took 20+ prompts to recreate in other tools, only required 2 prompts in Claude Design.

The implication for product teams isn’t “designers get faster.” It’s that the act of designing stops being a stage. Design becomes something that happens during the conversation where the decision is getting made.

This is the shift-left PM role, made concrete. PMs no longer hand off a spec and wait. They riff on the artifact alongside the designer and the engineer, in the same room, watching ideas resolve into pixels while the decision is still live. The feedback loop between “what if we tried…” and “here, look at it” closes from days to minutes.

Now, the tactics.

Ryan Mather’s seven tips for getting real work out of Claude Design

Ryan Mather is on the verticals team at Anthropic, which means he ships design work across seven different products. He posted a thread of the tips he’s using in practice. They’re worth reading as written, but here they are with a PM lens on why each one matters.

1. Set up your design system and your core screens. Ryan: “An hour of setup and refinement here is worth it.”

The temptation is to start prompting immediately. Resist it. Claude Design can read your codebase and design files during onboarding to build a working design system, and every downstream project inherits it. An hour spent refining tokens, components, and a few anchor screens compounds across every mockup your team makes for the next quarter. Skip this step and you’ll spend the rest of your week re-describing your brand to a model that already had access to it.

2. Iterate with your engineers live. Ryan: “I’m usually able to design new features with an engineer in a single meeting. Because Claude is super fast at mockups, we can stay high level in our conversation, riffing on concepts and constraints and watching them come to life.”

This is the tip with the biggest organizational consequence. The moment your engineers are in the room while the design takes shape, the “is this feasible?” conversation and the “is this the right thing?” conversation collapse into one loop. No more designs that die in engineering review. No more engineering estimates based on a mockup the engineer never helped shape.

3. Use the Comment tool for surgical edits. Ryan: “After a rough first draft, there might be dozens of details you want to tweak. It’s tricky to describe all the changes you want verbally — so don’t! Point and crit.”

PMs know this pain. You’ve typed out paragraph-long descriptions of a CTA adjustment that could have been solved by pointing at it. Inline commenting means the feedback lands where the decision lives. For teams doing async design review, this also means better review fidelity: reviewers comment on the thing, not a paraphrase of the thing.

4. Ask Claude to make video demos of your ideas. Ryan: “Claude Design can do almost anything you can think of. It’s honestly more like Claude Code than a canvas-based design tool.”

This reframes the whole product category. Claude Design is a code-powered prototyping environment that happens to be excellent at producing designs. A PM who grasps this can ship short video walkthroughs to stakeholders, interactive demos to sales, and working prototypes to user testing, all from the same conversation.

5. Use connectors, especially docs and Slack. Ryan: “Once set up, you can send prompts like ‘Please read the meeting notes from the product roast and create a deck exploring different design solutions for everything that came up.’ Go for a walk and touch some grass and come back with fresh eyes!”

This is where the shift starts looking less like “faster design” and more like “different workflow.” Connectors let Claude Design pull from the substrate where product decisions get made: the meeting notes, the Slack thread, the roast doc. The PM’s job stops being “translate this discussion into a brief” and starts being “point Claude at the discussion.”

6. Ask Claude to make bespoke on-the-fly tools. Ryan: “In general, don’t try to use Claude Design the way you would use a canvas-based tool. It’s a different animal with different superpowers. Experiment and get funky! You’ll find yourself designing circles around the way you used to.”

If you catch yourself replicating your old Figma workflow, stop. Claude Design can build you a custom slider to test spacing values, a side-by-side variant viewer, a small simulation of how your copy behaves at different lengths. The point isn’t to do old things faster. It’s to do things you couldn’t do before.

7. Know when to slow down and do things by hand. Ryan: “New icons, spot illustrations, naming. Some details will always make an outsized impact. It can be easy to get sucked into the hyper speed of agentic designing. Knowing when to slow down is an art form of its own.”

The most honest tip in the thread. Speed is a trap when the output is something that needs a human hand on it: naming a feature, picking the illustration that anchors a launch, getting the first icon in a new set exactly right. Claude Design gives you the speed. Your judgment decides when to deploy it.

Ryan’s summary of his own thread: “My favorite thing about Claude design is how delightful it makes the process of design! I can try out more divergent ideas, and hold them more loosely.”

That last line is the whole point. When the cost of trying a direction drops to zero, you try more directions. When you try more directions, you find better ones.

What this means for product development lifecycles

Zoom out from the tips. Three things change structurally.

Exploration becomes cheap. The default number of concepts a team considers for any given feature has always been bound by designer hours. Now it’s bound by PM imagination. If your team was shipping the first idea that looked fine, you were doing that because exploring five ideas cost a week. It doesn’t anymore.

Handoff becomes continuous. The traditional product-design-engineering relay had three batons: the spec, the mock, and the ticket. Claude Design’s handoff bundle to Claude Code means the mock and the ticket start merging. The spec follows. PMs who still write specs the old way, then mock separately, then write tickets separately, are going to feel slow against teams running all three through the same conversation.

Design review shifts from gate to dialogue. When design artifacts take weeks to produce, review is a gate: this passes or goes back. When they take minutes, review becomes a dialogue: what if we tried this, what about that. The product org that adjusts its review rituals to match will ship better work. The one that keeps gating will create a bottleneck in the one place the new tool didn’t.

Every customer signal. Zero manual tagging.

Book a demo

The MCP angle: what happens when product velocity meets live prototyping

Here’s where this gets pointed for product teams specifically.

Claude Design is great at turning ideas into mockups. It’s only as good as the ideas you feed it. The thing it cannot do, the thing no design tool can do, is tell you which idea is worth exploring in the first place.

That’s where an MCP like Bagel’s comes in. The Bagel MCP Server plugs an AI-Native Product Velocity Platform into Claude. A PM can ask, in natural language, what the top pain points are for enterprise customers this quarter, pull the underlying evidence with customer metadata attached, and narrow it down to a specific set of signals worth designing against.

Now picture the full loop.

You open Claude and ask Bagel’s MCP for the unaddressed pain points in your enterprise segment for the last ninety days. You get back a set of classified gaps with revenue-connected evidence. You pick the three worth exploring. You switch modes and ask Claude Design to mock up solutions for each, using your design system, pulling the exact customer quotes into the prototype copy. You invite an engineer to the session and refine the strongest direction live. When you’re aligned, you hand the bundle to Claude Code.

The signal drove the prototype. The prototype drove the build. No briefs lost in translation, no six-week cycle to find out the idea missed the actual pain point, no designing something beautiful that no customer asked for.

That’s not a workflow Claude Design alone can run. It’s not a workflow Bagel alone can run. It’s what happens when product velocity and agentic design sit in the same conversation.

What to do this week

Three things, in order.

Set up Claude Design properly. Give it an hour with your codebase and design files. Build the design system. Mock your three most-used screens as anchors. This is the tax that makes everything else fast.

Run one live design session with an engineer. Pick a feature you were going to spec next week and design it in a single meeting instead. See how it feels. Most PMs who try this don’t go back.

Start thinking about the connected loop. If your product discovery still lives in one tool, your design in another, and your customer signal nowhere, the compounding value of agentic design is bounded. The teams that wire these together early will compound faster than the teams that don’t.

Ryan ended his thread with a promise to share more tips as they emerge. We’ll do the same as we see how the PM role keeps shifting. The short version for now: the bottleneck isn’t production anymore. It’s knowing what to build.