Plan: becoming a top‑1% AI product manager¶

The radar is the input. Reading alone won't put anyone in the top 1%. What gets you there is a loop of read → understand → build → publish → get feedback, run every day for a year. This plan covers the system (what's built and what comes next) and the routine (what you do with it).

Part 1: The system¶

Phase 1: Daily radar (built, this repo)¶

Phase 2: Sharper signal (weeks 2–4)¶

Feedback loop. Add a 👍/👎 per story (a GitHub issue comment or a tiny form) and feed your votes back into triage as few-shot examples, so the radar learns your taste.
Direct X/LinkedIn. Turn on the Apify collector for about 30 key builders (lab leads, eval and agent practitioners, top AI PMs). Budget: about $5/month.
Story clustering. Merge five posts about the same launch into one story that lists all the sources.
Source health check. A weekly job that flags feeds which have gone silent or started failing.
Podcasts and YouTube. Pull transcripts of Latent Space, Dwarkesh, No Priors, Lenny's and lab keynotes, and summarize the PM-relevant minutes.

Phase 3: Knowledge you can query (month 2)¶

Searchable site. Publish digests, the knowledge base and the glossary with GitHub Pages (MkDocs), including search.
"Ask the radar". Retrieval over data/*.json so you can ask "what changed in agent evals since July?"
Trend tracker. Count category and keyword frequency over time and chart what's heating up.
Delivery. Email or Telegram push of the TL;DR at 7am (you already run a Telegram bot for pm-jobs).

Phase 4: Proof of skill (months 2–6)¶

Portfolio repo. One small shipped artifact a month, prompted by the weekend builds: an eval harness, an agent with guardrails, a model-routing cost calculator.
Teardown series. A monthly deep teardown of an AI product, covering its evals, failure modes, pricing and moat.
Talk. Turn the strongest quarterly thesis into a meetup or conference talk.

Part 2: The routine¶

Daily (35–45 min)¶

Time	Do
5 min	Read the TL;DR and trend signals
15 min	Read the 🔴 study stories in full at the source, and skim the rest
10 min	Work through the deep dive and answer the interview question out loud before opening the model answer
5 min	Flashcards (Anki, from `knowledge/flashcards.csv`)
5–10 min	Edit the LinkedIn draft: your hook, one line from your own experience, then post

Weekly (Sunday, 90 min)¶

Read the weekly review and write your own one-paragraph "what I now believe".
Do the weekend build (≤2 hrs). Shipping something small beats reading something big.
Publish the long-form post.
Pick one 🔴 paper or engineering post to read end to end and annotate it in notes/.

Monthly¶

Re-read your posts and track which ideas resonated. That's your emerging point of view.
Score yourself against CURRICULUM.md (1–5 per competency) and aim next month's study at the lowest score.

Part 3: LinkedIn strategy¶

The goal is a recognizable point of view, not daily noise. 1. Pick two pillars and own them (suggested: AI evals & quality and agentic product design). About 70% of posts go here, 30% on the frontier generally. 2. Formats that compound: "what X's launch means for builders" (fast take, same day), "I tried X, here's what broke" (hands-on), "framework" posts (from the deep dives), and monthly teardowns. 3. Cadence: 4–5 posts a week, never pure summaries, and always one opinion plus one concrete takeaway. 4. Engagement: spend 15 minutes a day leaving substantive comments on posts by the people in sources.yaml. It grows the network faster than posting. 5. Measure: log impressions, comments and follows per post in a sheet, then double down on the topics and formats that work.

Part 4: Success metrics (12 months)¶

365 briefs read, 200+ posts published, 50 weekend builds, 12 teardowns
Able to whiteboard: an eval suite for an agent, a governance and risk plan for a launch, the cost/latency/quality trade-off for model routing, and an AI product strategy against frontier-model commoditization
Inbound: recruiters, founders and speaking invites citing your posts