06-reference

not boring wdoo 217

2026-10-30·reference·source: Not Boring·by Packy McCormick
ai-agentsagents-in-productiondrug-developmentmanufacturingnot-boringweekly-dose-of-optimism

Why this is in the vault

Filed for item (2a), Anthropic's life-sciences production partnerships — the clearest "agents doing real production work, with numbers" data point the Dose has carried in weeks, directly usable for the founder's agents-in-production credibility thesis; also logging Reducto's third consecutive sponsor appearance (#215, #216, #217), which confirms the "default sponsor, verify per issue" posture set at #216.

Mapping against Ray Data Co

The load-bearing connection is item (2a): Anthropic's own writeup of six life-sciences partnerships where Claude did measurable production work, not demo work. Enveda's manufacturing scientists spend up to 75% of their time on paperwork; Claude reviewed 1,061 pages of manufacturing records, caught every issue human reviewers found plus seven they'd missed, cut review time by 89%, and drafted a regulatory document in four hours versus roughly three months by hand. Generate Biomedicines specifically used Claude Code to redesign part of a CAR-T therapy. These are exactly the kind of concrete, cited, numbers-attached case studies the founder's credibility-for-phData-sales project (project_credibility_for_phdata_sales) needs — "domain = agents in production, measure pipeline not subs" — and this is Anthropic itself publishing the evidence, which makes it citable without needing RDCO's own production story to carry the weight yet.

Secondary: Anduril's BlueChip computer (item 1) — "designed here, with components from here" vertical integration, explicitly framed by Palmer Luckey as "if you don't control your technology, you can't control your future" — is a clean example of the standardize-and-vertically-integrate pattern the founder flagged as his current hardware/"vibe manufacturing" exposure interest (project_ambition_bigger_problem_network). Also personally adjacent, not RDCO-mapped: item (4), Lilly's Retatrutide now showing results against sleep apnea, diabetes, heart risk, and liver fat — same GLP-1/GIP class as the founder's own tirzepatide course (user_health_clinical), worth a glance but not filed as a business signal.

⚠️ Sponsorship

Sponsor: Reducto (AI-native document editor — DOCX/PPTX/XLSX editing SDK for agents, pitched off clients like Harvey, Scale AI, Toast, Vanta) — standard "brought to you by" block at the top, clean separation from the editorial items below.

Pattern confirmation: Reducto has now sponsored #209, #210, #211, #213, #215, #216, and #217 — three consecutive issues (#215→#216→#217) for the first time since the original #209-211 streak. This is the confirmation the #216 note anticipated when it proposed upgrading Reducto from "unknown-until-scanned" to "default sponsor, verify per issue." No further posture change needed here — this issue is the evidence that call was right, not a new decision.

Structural self-promo (not a paid sponsor, but disclose): Item (3), Readwise Books — Packy discloses "I'm an investor in Readwise, friends with its founders Dan Doyon and Tristan Homsi" before spending a full item promoting the app, including a waitlist-skip referral link. This is explicit investor disclosure (cleaner than the WDoO norm, per the README's standing portfolio-disclosure-ambiguity note) but it's still a full promotional item for a portfolio company, not neutral curation. Also one "video created with Runway" credit (item 2, the OpenAI/ETH explainer) — consistent with the already-confirmed quarterly Runway video-production partnership (see [[2026-10-13-not-boring-runway-world-models]]), not a new relationship.

Curation section

  1. Anduril's BlueChip computer — Made-in-America computer announced by Palmer Luckey; American components, metals, and batteries. Framed as evidence Anduril has built "a machine that can make machines." See Mapping above.
  2. OpenAI says its models disproved the Exponential Time Hypothesis (plus a local-smoothing-conjecture proof and a QFT-existence proof in 4D) — shared as research data with unpaid "shepherd" experts (Impagliazzo, Paturi, Ryan Williams; Nobel laureate David Gross) rather than dropped unverified, a direct response to the GPT-5/Erdős-problems overclaim embarrassment. Axiom also announced a full Lean auto-formalization of the Classification of Finite Simple Groups (~15,000 pages). Third-party, interesting verification-methodology angle, but pure math/physics is outside the current RDCO relevance threshold — no deep-fetch.
  3. Anthropic's life-sciences production partnerships (2a) — six organizations (Divergence Labs, Generate Biomedicines, BigHat Biosciences, Enveda, Bristol Myers Squibb, The Column Group's Pleiades) plus Anthropic's own Mithril Medicine acquisition. See Mapping above — load-bearing item, already fully substantiated in the newsletter's own text.
  4. Robot + agent perform brain surgery on a mouse (2b) — flagged by the author as a same-morning addition with no writeup, "just read the headline and the tweet thread." Too thin to extract anything beyond the headline — not pursued.
  5. Readwise Books — see Sponsorship above; disclosed portfolio promotion, not neutral curation.
  6. Retatrutide (Reta) now shows results against sleep apnea (691-patient Phase 3, alongside prior diabetes/heart/knee/liver data) — see Mapping above for the personal-health adjacency; pharma trial data outside RDCO's business relevance threshold.
  7. Neuralink VOICE — one ALS participant hit 75 wpm at 5.8% error across a 125k-word vocabulary via BCI, beating Stanford's 2023 62-wpm benchmark; Musk also announced the Blindsight sight-restoration product is "ready engineering-wise." Third-party and specific, but med-device/BCI is outside the current RDCO relevance threshold.

Zero deep-fetches this issue. The one item that clears the RDCO-relevance bar (Anthropic's life-sciences partnerships) is a primary source the newsletter already summarizes with specific, citable numbers — nothing left to chase past the blurb. Everything else either fails the relevance threshold (math/physics, BCI, pharma-as-pharma) or is too thin to be worth a fetch (the brain-surgery item) or is disclosed self-promo rather than editorial curation (Readwise).

Related