How to Vibe a PM Agent
Automating the PM Role: Warp + MCP + Linear in Action
PMs and Engineers are loving “designing” through Vibe-Coding and creating “UX prototypes”… well, it’s about time designers “vibe-pm’ed”.
That’s exactly what I explored in this demo using Warp, an MCP server, and Linear. The experiment: write a single prompt, let Warp generate a Product Requirements Document (PRD) for a simple Next.js weather app, and then watch as it auto-creates a full backlog of tasks inside Linear.
The Setup
Warp provides a clean, diff-first command line interface (no clutter, every change is transparent).
MCP bridges to Linear, so the agent can create projects, epics, and issues directly.
A basic prompt kicks things off: “Outline a Next.js weather app as a PRD and prepare a task list to post to Linear.”
The Run
The workflow unfolded step by step:
PRD Generation – Warp produced a structured PRD, including features, tech stack, and user stories.
Backlog Creation – The agent split the PRD into three epics with tasks for each, complete with acceptance criteria and definitions of done.
Linear Integration – Everything landed in a brand-new Linear project, neatly organized and ready for work.
Warp’s diff-first UI made the process surprisingly transparent. Switching Auto mode on allowed the agent to execute task after task without interruptions.
Lessons Learned
Scope Control: A basic prompt creates a lot. Add terms like “lean/minimal” to keep things digestible.
Testing Early: Always prompt for test scaffolding (e.g., Playwright for UI apps). It saves headaches later.
Token Awareness: Long PRDs + task lists can eat tokens quickly—worth keeping an eye on.
Limits: This is great for spinning up MVPs and prototypes, but it won’t replace the strategic and organizational value a human PM brings.
Why This Matters
Instead of spending hours drafting PRDs and writing Jira/Linear tickets, you can go from idea → backlog in minutes. It’s not about replacing PMs, but about reducing the overhead of starting from zero. Teams can begin prototyping faster, experiment more, and iterate without getting bogged down.
In the end, I had a live Linear project with a PRD, epics, and tasks—all from one command-line prompt. And from there, you can pass it directly to a coding agent to start building.
This is what “agentic workflows” look like in practice: designers and engineers equipped with lightweight AI PMs, embedded directly in their tools.
