Cycle on Sep 22, 2026

I'll start by reviewing the open threads and existing notes for this topic, then drill into the HN stories that touch the brief (Panel — a research workspace — and "AI is breaking our proxies for expertise" both look relevant).

The topic has a rich existing note set. This cycle's HN map has two directly relevant leads: **Panel** (a research workspace — direct lookalike for Hey, Lefty, with an existing note revised yesterday) and **"AI is breaking our proxies for expertise"** (relevant to the value-prop argument). Let me drill into both stories.

I'm mid-cycle: I've reviewed threads and notes, and drilled both HN leads (Panel, Goodhart essay) — but both appear already digested by existing notes. My next step is to fill a genuine gap: the academic "Deep Search startup" layer (Undermind, Ai2 Asta) named in the Scholar Labs watch but lacking any note, plus a quick check of the Scholar Labs watch status. Running those searches now.

Good leads. Scholar Labs shows **no GA news** (watch stays open), but two significant items surfaced: Google just launched **"Expert Intelligence" in Gemini Notebook** (licensed publisher content — a data-moat move for the notebook layer), and there's substantive material on Undermind and Ai2 Asta — the academic Deep Search layer that has no note yet. Fetching the primary sources now.

The Undermind guide is substantive (pricing, GSK deployment, Semantic Scholar index dependency). The Google blog fetch returned only navigation chrome — I need the Expert Intelligence content from another angle, plus the primary Asta announcement.

The Asta announcement is rich (MCP-first Scientific Corpus Tool, AstaBench, open-source agents). Scholar Labs shows no GA news — its watch stays open. Now let me get the primary source for Expert Intelligence (the Google blog post) for verbatim quotes.

Primary source confirmed. I have everything needed. Writing the two findings now: (1) the academic Deep Search layer (Undermind + Ai2 Asta) — a genuine gap in the market map; (2) Google's Expert Intelligence launch — the licensed-source supply play in the notebook layer.

The first note landed; the second call was sent empty by mistake. Writing the Expert Intelligence note properly now.

Both findings are persisted. Now recording the open question (Undermind's unverified recall claims) and registering the watch on Expert Intelligence's expansion roadmap.

Cycle complete. Summary of what this cycle did: **Verified, didn't rewrite:** The cycle's two HN leads — Panel's Show HN thread and the Goedecke "proxies for expertise" discussion — turned out to be already digested by existing notes (`agentic-research-workspaces-panel`, `ai-expertise-proxies-goodhart`), so I avoided redundant re-minting. **Two new findings (market-map gaps filled):** 1. **`academic-deep-search-undermind-asta`** — the academic Deep Search layer now has a complete price ladder: Undermind (YC S24, $0/$16/$15/custom tiers, GSK's 1,000-scientist "AI Scientist" deployment, enterprise backend ambitions) vs. Ai2's free, open-source, MCP-first Asta (200M-paper corpus tool, AstaBench). Key implication: commercial players are bracketed between Google's free Scholar Labs and a non-profit giving away the same Semantic Scholar index — differentiation must come from the agent loop and enterprise workflow, not data access. 2. **`google-expert-intelligence-licensed-sources`** — Google's Aug 27 launch of Expert Intelligence (100k+ licensed publisher books in Gemini Notebook, ownership-gated via Play Books) is a licensed-source supply-chain moat, with a roadmap explicitly pointing at "third-party subscriptions, business research reports, and textbooks" — the exact supply premium research-data vendors sell. **Housekeeping:** Recorded one open question (independent verification of Undermind's recall claims) and one tripwire watch (Expert Intelligence expansion beyond books). Checked the Scholar Labs GA watch — no news, stays open.

The research cycle is complete. This cycle extended the market map with two new findings — the academic Deep Search layer (Undermind vs. Ai2's free Asta) and Google's Expert Intelligence licensed-sources play — plus one open question and one forward-looking watch, while confirming the cycle's HN leads were already covered by existing notes.