A PM's guide to building working prototypes with AI
Replace the PRD with a behavioral brief, generate an interactive prototype, iterate in the same environment, then hand engineering a tested prototype.
- Replit
34 workflows Updated
For PMs, designers, researchers and design ops. Each entry lists the steps and tools, and some include a prompt you can copy.
34 workflows
Replace the PRD with a behavioral brief, generate an interactive prototype, iterate in the same environment, then hand engineering a tested prototype.
Plan first in Bolt Plan Mode or Lovable Chat Mode, then build one chunk at a time with precise design vocabulary.
The same project built four ways with v0, Lovable, Bolt and Cursor, plus a decision tree for which to use.
A comparison of Lovable, v0, Bolt, Replit, Cursor and Magic Patterns for designers and PMs who prototype.
Plan with the AI, build on a branch of your actual product, preview in context, then turn the validated prototype into a pull request.
Set up Figma's hosted MCP server in Cursor, Claude Code, Codex or VS Code and generate code from selected frames.
Structure files with components, variables and Code Connect so agents stop guessing, and trigger the right MCP tools.
A rules file that forces the agent to fetch context and a screenshot before coding and to reuse your design system tokens and components.
Use the Figma MCP server in Cursor to read components and variables, generate code, and send live UI back to Figma as editable layers.
Code each transcript on its own, then cluster codes into themes so every theme keeps a participant ID and verbatim quote.
One file per interview, a CLAUDE.md with integrity rules, themes with verbatim quotes, and a final verification pass.
A Claude skill that turns transcripts into a synthesis report plus a separate problems-only document, with quote verification.
Daily routines for research synthesis, personas, usability analysis and design critique using Claude Code skills.
Paste raw notes into Claude and get 3 to 5 themes with quotes, interpretation, implications and flagged contradictions.
A gated prompt that builds a PRD section by section, labels gaps as assumptions or open questions, and ends with a self-critique.
A looped agent goal that drafts PRD sections in dependency order with sign-off per section, writing the executive summary last.
Use @-mentioned context, Socratic questioning, parallel strategic options and sub-agent reviews before sharing a PRD.
Copy-paste prompts for problem statement, user stories, scope, metrics, edge cases, risks and summary, plus a template.
Run component audits, token consistency checks, bulk renames and layer cleanup with an agent inside the file.
Prompt Figma's agent to write component specs, annotations and handoff notes from what is actually on the canvas.
Use the use_figma tool so an agent creates real frames, components, variants and variables in your file.
Spec Layer documents components and tokens on canvas and exports YAML and DTCG specs that a repo can pull via CLI.
Use AI as a pre-review and analysis layer for usability sessions while keeping humans on validation.
Pair automated accessibility checks with agent-run keyboard, focus and screen reader checks and human sign-off.
Non-coding uses of Claude Code collected from Lenny's community, many for PM work.
Batch-process feedback CSVs in Claude Code, require 3 supporting rows per theme, and keep row IDs and unclustered items.
Teresa Torres on research, writing, context libraries and a custom to-do system with Claude Code.
Dovetail agent workflows turn tagged customer feedback into Linear issues with quotes and account context.
Linear suggests teams, assignees, labels, projects and duplicates for incoming issues from content and history.
Notion AI transcribes and summarizes meetings and runs agents for recurring work across connected apps.
The Figma MCP server's generate_diagram tool turns Mermaid from your agent into an editable FigJam diagram.
Constrain agent critique with a brief, evidence packet, severity levels, and a read-only findings pass before any revision.
Write a SKILL.md that tells coding agents to search your library first, bind tokens, and reuse real components instead of inventing UI.
Define one release-critical journey, run a simulated participant through it, and keep observed facts separate from interpretations and hypotheses.
No workflows match these filters yet.
One email a week with what was added to the directory.