Launching today

Lumen — Run PM Workflows with AI Agents

Launching today

1 Orchestrator. 18 AI Agents. 6 PM Workflows. In Claude Code

10 followers

1 Orchestrator. 18 AI Agents. 6 PM Workflows. In Claude Code

10 followers

Visit website

AI Chief Of Staff

Lumen is a Claude Code plugin that orchestrates 18 specialized AI agents across 6 PM workflows — PMF discovery, churn recovery, quarterly strategy, feature validation, GTM launch, and NRR analysis. Free to start. Open source MIT license. Built on Claude Agent SDK.

Lumen — Run PM Workflows with AI Agents gallery image

Free

Launch tags:Open Source•Developer Tools•Artificial Intelligence

Launch Team / Built With

Framer — Launch websites with enterprise needs at startup speeds.

Launch websites with enterprise needs at startup speeds.

Promoted

MockRabit

Maker

📌

Hey there! Throughout my journey, and in conversations with dozens of PMs, the same frustration kept coming up. We know the frameworks, but executing them rigorously is hard. PMF scoring, opportunity trees, NRR decomposition, and launch readiness audits require discipline, yet they often get skipped, rushed, or replaced by gut calls. Lumen is how I encoded 30 years of PM practice into 18 AI agents. Here's what makes it different from every other "AI for PM" tool you've seen: It's not a SaaS. There is no login, no dashboard, and no subscription required. It runs as a Claude Code plugin in your terminal, where you already work. One command and a structured PM workflow begins. It's designed and architected to provide superior results Each of the 18 agents is a `.md` file with YAML frontmatter. Want to change how PMF scoring works? Edit `agents/signal-monitor.md`. This is the "agents-as-code" pattern pushed to its logical conclusion. Every recommendation has an evidence grade. HIGH / MEDIUM / LOW. The report tells you exactly which sections have strong data backing them and which are directional guesses. It calls out the missing data by name. No hiding uncertainty. The AI asks permission before irreversible decisions. For Pricing changes, segment pivots and feature deprecations, Lumen pauses and presents a structured approval gate. You type `APPROVE`, `DECLINE`, or `MODIFY`. The Lumen AI agent doesn't move without you. It starts free and degrades gracefully. No PostHog? PMF scoring is skipped but the rest of W1 runs. No Supabase? Knowledge Graph writes are silently skipped. Missing an MCP? The workflow continues, flags what's missing, and carries on. No hard failures. The agent is free to use. We have open sourced it for you to customize. A few things I'd love to hear from you: - Are you using Claude Code? Would a PM co-pilot that lives in your terminal be useful to you? - What PM workflow gives you the most pain? Discovery? Churn diagnosis? Launch readiness? The MIT licensed GitHub repo is linked. Star it if you think this matters to you as a PM, your support will help me know the signal that the community believes in this direction. — Ishwar

Report

3d ago

MockRabit

Maker

Sharing one thing I learned building this: the hardest part wasn't the agents. It was designing the evidence quality aggregation system. You can't just average confidence scores — a LOW from a primary agent should sink the whole report. But a skipped agent (e.g., SignalMonitor when PostHog is disconnected) shouldn't count as LOW. Took 3 iterations to get that right.

Report

6h ago

Forum Threads

p/lumen-run-pm-workflows-with-ai-agents

•

2d ago

What's the most expensive PM decision you've made on bad data?

I've spent 30 years in product. Built and launched close to 115 products for companies like Sony Music and Zee Entertainment where I worked, for my own company and for companies that I consulted.

The most common pattern I've seen kill good products isn't bad strategy. It's decisions made with false confidence where the data looked complete but wasn't, and nobody said so out loud.

View all