What is this?
The Sharp PM is an experimental project for Product Managers who want to stay sharp without drowning in newsletters. It does two things: a daily aggregator that pulls fresh content from PM and AI sources, and Agent Sharp, a weekly editorial dispatch with a point of view.
Each piece in the daily digest is summarised and tagged by AI, so you can scan, filter, and skip what you don't need. Agent Sharp goes further: it reads the week's output, picks what matters, and writes an opinionated editorial.
How it works
Two pipelines run on GitHub Actions. The daily aggregator fetches RSS feeds, sends each item through an OpenRouter-served model that returns a 2-3 sentence summary and topic tags, and writes the result to a JSON file. The site is a static page that reads that JSON.
Agent Sharp runs every Sunday. It's an agent loop: the LLM reads its own past editions, browses the feeds, searches the web via Tavily, fetches full articles when something catches its eye, and finally publishes a single opinionated dispatch. Strict gates on the publish step refuse fabricated URLs, untrusted sources, meta-narrative prose, lazy descriptive "why" fields, statistics that appear in no source, pull quotes that don't match the article verbatim, and themes that repeat a recent edition.
Every run is traced in Langfuse — one span per model call and per tool call, scored on whether it published and how many times the gates sent it back. That's how the gates get calibrated: the thresholds come from what the archive actually did, not from guesswork.
No server. No database. No ongoing costs.
- Hosting GitHub Pages
- Automation GitHub Actions
- Daily LLM OpenRouter (GLM-4.5-Air)
- Agent LLM Groq (gpt-oss-120b)
- Web search Tavily
- Tracing Langfuse
- Content RSS feeds
- Built with Claude (Anthropic)
Sources
Contact
Feedback, suggestions, broken links, sharp takes — all welcome at thesharppm@gmail.com.