Claude Code for Product Managers: Prototype Without Engineering

The PM tax has always been waiting for engineering bandwidth to validate something. A feature flag, a quick prototype, a hacky internal tool — you write the spec, file the ticket, wait two weeks, ship a worse version of what you actually wanted because the engineer ran out of time.

Claude Code, used from a visual canvas, eliminates that cycle for a real subset of PM work. This guide covers what you can actually prototype yourself, how to do it, and where you should still pull in your eng team.

What PMs can prototype solo with Claude Code

Real, working prototypes you can put in front of users today, no eng involvement:

• Functional clickable prototypes — not Figma clickthroughs, actual web apps with state, routing, mock APIs
• Internal tools — quick admin dashboards, data viewers, CSV transformers
• User research artifacts — landing pages for fake-door tests, survey microsites, onboarding flow tests
• API mock servers — for testing client team integration before backend is real
• Customer-facing pages — pricing pages, feature pages, comparison tables
• Quick analyses — pull data from a CSV, transform, visualize, share

These all share a property: they're scoped, they don't need to integrate with your production systems, and they need to be in someone's hands within a day or two.

What you can't (and shouldn't) try to solo prototype

Things that still need your eng team:

• Anything that touches your production database
• Anything that requires deploying to your production infrastructure
• Anything that handles real customer data or PII
• Anything that needs to integrate with your auth system
• Anything that you'd expect to maintain for more than a sprint

Claude Code can technically do all of these. But your security, infra, and codebase conventions matter — that's why your engineers exist. Claude is for the "I need to validate this idea this week" layer.

The setup

Same as the other Claude Code workflows:

1. Mac. Required.
2. Anthropic Claude Pro or Max subscription ($20–$200/month).
3. A canvas surface to drive Claude Code from forms — using Clearly here. Free to start.

About 15 minutes of one-time setup, then every prototype is ~30 minutes to first preview.

Workflow: "I want to validate this feature this week"

1. Write a short brief

Not a PRD. A short paragraph in the form on your canvas:

"Internal tool for the customer success team to look up users by email and see their last 10 events. Auth via password (one shared one is fine). Should run on a single page. Use mock data — the CS team will tell me what real data they want once they see the shape."

The "Internal tool" template on Clearly has a form that prompts for: name, audience, primary action, data shape, auth requirements, deploy preference. Fill that in.

2. Submit, watch it build

Status block updates as Claude scaffolds the React app, sets up routing, writes the lookup component, generates mock data, deploys to a preview URL.

A simple internal tool: ~5 minutes. A multi-page prototype with state: ~15 minutes.

3. Get the link, send to the CS team

Preview block populates with the URL. You paste it in Slack. CS team plays with it. They tell you what's wrong.

4. Iterate from the same form

You re-open the form on the canvas, add a quick note ("add a 'last login' column and let CS edit a user's plan tier"), submit. Claude reads the existing project and modifies it. New preview in 90 seconds.

5. Decide

After two or three iterations the CS team either says "this is exactly what we need" (you hand it to eng to build properly, with a working spec) or "actually we don't need this" (you've saved them a sprint of waste).

The whole cycle: two days, zero eng involvement.

A second example: fake-door landing page

A PM wants to test demand for a feature before building it. The classic move is a "Coming Soon" landing page with an email capture, count signups, decide.

Old way: file a marketing ticket, wait for design + dev capacity, ship in 2 weeks, test for 1 week, decide.

New way:

1. Drop the "Landing page" template on a canvas.
2. Fill out the form — value prop, audience, CTA copy, brand colors.
3. Claude ships the page in 5 minutes, deployed to a preview URL.
4. Set up the URL as a custom domain (5 minutes if you have a domain).
5. Drive traffic from your normal channels.
6. Track signups (Claude wires up a simple emails-to-Notion or emails-to-CSV flow).
7. Decide.

Total elapsed: half a day. Cost: $0 incremental.

A third example: API mock for a client team

Backend engineer says "the new API will be ready in a sprint, here's the schema." Frontend team is blocked.

PM moves: drop the "Mock API" template on a canvas, paste the schema into the form, submit. Claude scaffolds a Node + Express server that returns realistic mock data matching the schema, deploys it to a preview URL. Frontend team builds against the mock until the real API ships.

Two prevented weeks of blocked work. Zero engineering involvement on the PM's side.

Why a canvas matters specifically for PMs

PMs work cross-functionally. A canvas you can share with stakeholders is a different artifact than a terminal session you alone can see.

Specifically:

• Status feed. Stakeholders watching the canvas see "Claude is writing the lookup component" instead of getting impatient about progress.
• Shareable previews. Customer success team opens one link, sees the prototype, leaves comments directly on the preview block.
• Multi-project organization. Each prototype lives on its own canvas. You can pull up "the prototype I built for marketing last month" instantly.
• Brief lives with the result. When you come back in 3 months, the form you filled in is right there next to the deployed URL. Self-documenting.

A terminal can't do any of this. Lovable and Bolt do some of it but hit the complexity ceiling fast on PM use cases.

What it costs

• Clearly Free: 1 canvas, 3 active projects. Fine for trying.
• Clearly Pro: $19/month — unlimited prototypes, share with stakeholders, full template library.
• Anthropic Claude Pro: $20/month.
• Total: ~$39/month.

If this eliminates one "wait for eng bandwidth" cycle per month, it's paid for itself ten times over in PM time.

When to use this vs. when to file an eng ticket

Use Claude Code on canvas	File an eng ticket
Throwaway prototype	Production code
Internal tool with mock data	Tool with real customer data
Landing page for a fake-door test	Conversion-critical funnel page
API mock for unblocking frontend	Real API connected to real DB
Customer-facing static page	Integration with your auth system
Quick data transformation	Data pipeline that runs forever

Start your first PM canvas → — first prototype in under an hour.