{ "flags": [ { "id": "f1-skills-failure-test", "meeting_id": "NPM16-04", "kind": "worth_spreading", "title": "Likely Person B says nobody cares about skills; Person D would still ship them", "face": "In Wednesday’s Agent product review, Person D proposed turning automations into skills that run on a schedule. Likely Person B objected: “I don't think anyone cares about skills,” and usage data backs that up. Likely Person D didn’t dispute it. He set a failure test—skill usage rises but satisfaction, revenue and memberships don’t—then asked: if the product had no skills today, would we ship them? His answer: yes, as a way to deliver value.", "context": "Person D was walking Person B through the Agent roadmap, where the first step is connecting automations to skills (“automations become basically skills on a schedule”). Person B, who wants roadmap work tied to metrics that will really move, asked whether skills matter to users at all.", "moment": [ { "line": 99, "speaker": "likely Person B", "quote": "So far, maybe a hot take, I don't think anyone cares about skills. Outside of maybe us, right?" }, { "line": 106, "speaker": "likely Person D", "quote": "If we ship a bunch of updates to Skills and then Skill usage goes up because we like, you know, orient around that. But. like overall satisfaction, revenue. Memberships? Don't change. then that's a failure." }, { "line": 108, "speaker": "likely Person D", "quote": "if we had no skills in the product right now, would we ship skills? And I think this sort of outlines like, Yes." } ], "why": "It gives you two clear tests: judge skills by satisfaction, revenue and memberships, and ask whether you would build them from scratch today. That goes beyond July’s finding that users mostly consume skills rather than build them. This is a discussion, not a decision; skills were not removed, and the transcript has no speaker labels.", "strategy_quote": null, "reaction_inferred": null, "options": [], "people": [ "Person B", "Person D" ], "confidence": 0.72 }, { "id": "f2-never-repeat-yourself", "meeting_id": "NPM16-01", "kind": "worth_spreading", "title": "Compounding looked like homework on the Agent page; the fix: “never repeat yourself”", "face": "Reviewing the Agent landing page, likely Person I said the compounding section “makes me feel like I have to do all of these things”; he just teaches the agent something once. Likely Person D replied that “the compounding happens automatically,” and the room liked “never repeat yourself” as a header, next to your “don’t make the same mistake twice.” That afternoon, Person D’s proactive skill-creation roadmap item was tied back to that phrase.", "context": "At the launch standup, Person H walked through the Figma design for the Agent landing page. The blue “watch it compound” section showed a chain of teammates correcting the agent. The group was deciding how much to emphasize compound engineering and compound writing on the page.", "moment": [ { "line": 113, "speaker": "likely Person I", "quote": "I think one problem that I have with this section is It makes me feel like I have to do all of these things." }, { "line": 117, "speaker": "likely Person D", "quote": "The agent learns how you work. So that next time you ask it to do something, it does it the right way. And so the compounding happens automatically, but it's not literally compound engineering every time working in the agent." }, { "line": 118, "speaker": "likely Person E", "quote": "I wonder if even never repeat yourself or something might be a nice header." } ], "why": "Earlier cards covered pushing compounding into the future; this one gives a concrete wording fix: describe the benefit (you never repeat yourself), not the method. In that afternoon’s product review (NPM16-04 L116), proactive skill creation was explicitly linked to “the compounding don't repeat yourself idea,” so the phrase is shaping the roadmap as well as the copy. The header is not final.", "strategy_quote": null, "reaction_inferred": null, "options": [], "people": [ "Person I", "Person D", "Person E", "Person H" ], "confidence": 0.66 }, { "id": "f3-uc-davis-appropriate-risk", "meeting_id": "NPM16-05", "kind": "room_texture", "title": "UC Davis IT director put the Agent in a 30,000-person Slack: “appropriate risk for now”", "face": "In a beta onboarding Wednesday, Person M, executive director of IT for UC Davis’s College of Engineering, installed the Agent in a Slack workspace he said has about 30,000 people. The hosts called it likely the largest install so far. He wanted private channels first and connected no legally protected repositories. He said: “I am in a position of taking on risk for my college and this is an appropriate risk for now.”", "context": "Person D and Person B were watching Person M, a longtime subscriber who uses Cora and Monologue daily, go through Agent setup. He talked through the university’s data protection levels: FERPA student data, defense data and California privacy law. He also said he may later move the install to a personal workspace.", "moment": [ { "line": 84, "speaker": "Person M (inferred from Q&A continuity)", "quote": "I would want to add it to private channels first. Um, We have about 30,000 people in the Slack workspace." }, { "line": 94, "speaker": "Person M (inferred from Q&A continuity)", "quote": "But I am in a position of taking on risk for my college and this is an appropriate risk for now." }, { "line": 115, "speaker": "Person M (inferred from Q&A continuity)", "quote": "I will invite it to a channel with a few of my peers and we'll play around and then grow it from there." } ], "why": "This is a first-hand look at the AI-ambassador buyer at the risky end: one person personally owns the risk in a large, cautious institution and wants a small private start, while onboarding assumes public channels. Separately, Person D’s Slack outage thread (SPM16-04) mentions a new ~30k-member workspace failing directory syncs. That may be this install, but it is unconfirmed, and Person J attributed the restart itself to his env-var change.", "strategy_quote": null, "reaction_inferred": null, "options": [], "people": [ "Person M", "Person D", "Person B" ], "confidence": 0.62 }, { "id": "f4-priming-disclaimer", "meeting_id": "GPM16-03", "kind": "worth_spreading", "title": "Telling the agent to hide counts made it drop numbers; the disclaimer scored worst", "face": "Engineer A’s open Plus One PR tried telling the agent not to repeat a provider’s internal counts, like “112 candidates, kept 5.” On an unrelated clarity test that scored 3/3 on main, timings like 250ms started turning into vague words, dropping to 1/3. Adding “never overrides the numbers rule” scored 0/3. The PR’s conclusion: “Priming does not read disclaimers.” Ban the specific artifact, not the category.", "context": "PLUSONE-1610: an external copy-edit job returned success while one of its stages had crashed. The agent neither retried nor translated the provider’s diagnostics, and the user replied “i dont know what you're talking about.” The fix adds a retry-once branch and a plain-language rule, and the vocabulary half went through four measured wordings.", "moment": [ { "line": 70, "speaker": "Engineer A (PR author account; body generated with Claude Code)", "quote": "every grader reported the same failure: the planted timings (250ms, 120ms, 8s) dissolved into qualitative phrases **in replies where nothing had failed and no tool had been called**." }, { "line": 75, "speaker": "Engineer A (PR author account; body generated with Claude Code)", "quote": "This scored **worst (0/3)**. Priming does not read disclaimers; a reassurance names the class as loudly as a ban." }, { "line": 355, "speaker": "Engineer A (PR doc, same PR)", "quote": "Before you add a prohibition to a prompt, ask what *class of thing* it names." } ], "why": "It is a counterintuitive, reusable lesson for anyone writing agent or skill instructions, including your writing rules: a “don’t do X, except…” clause can spread X-avoidance to unrelated work. Weigh it lightly: each wording was tested only three times, the PR is open and says not to merge before a clean sweep, and the delete-the-words fix was still being measured.", "strategy_quote": null, "reaction_inferred": null, "options": [], "people": [ "Engineer A" ], "confidence": 0.58 } ], "reading_notes": [ { "meeting_id": "NPM16-01", "substance": "Agent launch standup: Person H walked through the landing-page design. The group debated the skills section, the compounding section, a permissions example, and pricing (free to start vs. a trial; possibly dropping enterprise). Person D said billing and beta emails were planned for tomorrow.", "selection_reason": "Selected: the compounding-feels-like-homework to “never repeat yourself” reframe. The pricing talk is still a whiteboard proposal, so it was skipped." }, { "meeting_id": "NPM16-02", "substance": "Beta onboarding with Person K, a ghostwriter whose Slack is mostly clients. She saw the Agent as a team product, asked about credits and a Claude connection, accepted a call-prep automation, and responded to the host’s client-ops pitch as an alternative to hiring a VA.", "selection_reason": "Skipped: a real reaction, but the host pitched project management, not the AI-ambassador message, and the result of the automation isn’t known yet." }, { "meeting_id": "NPM16-03", "substance": "Person L and Person E planned the Sept 25 All Access office hours. Person E gave the “one company agent AI-pills the org” pitch and said he had 30–40 examples. Person L said he couldn’t find Person I’s Remotion skill by asking the agent, and that getting answers from Person B and Person D is hard.", "selection_reason": "Close runner-up (skill discoverability versus the show-don’t-tell pitch). Left out to keep skills from dominating the batch. Person E’s pitch was to a teammate, so it doesn’t count as message_tested." }, { "meeting_id": "NPM16-04", "substance": "Agent product review: Person D covered automation-first activation, turning automations into skills on a schedule, proactive skill creation, and better suggestions. Likely Person B challenged whether skills matter and argued against fake focus. They also discussed spec detail and starting retros.", "selection_reason": "Selected: the skills failure test and would-we-ship-it-today exchange. The <20% activation number was not resurfaced." }, { "meeting_id": "NPM16-05", "substance": "Beta onboarding with Person M, UC Davis Engineering IT director: an install in a Slack of about 30,000 people, private-channel-first wishes, data-protection levels, and a personal risk judgment.", "selection_reason": "Selected as outside-person texture from what the hosts called likely the largest install. The possible link to the afternoon outage is marked unconfirmed." }, { "meeting_id": "SPM16-01", "substance": "Person N’s one-prompt golf and sourdough games impressed an external EAP contact (“This is WILD”) and Person P.", "selection_reason": "Skipped: fun, but private early-access model reactions don’t support a broader claim." }, { "meeting_id": "SPM16-02", "substance": "Person Q’s writing spot checks: early-access models’ intros lack coherence, social pitches are navel-gazey even with the Reader A lens skill, and an email draft leaves out context readers need.", "selection_reason": "Skipped narrowly: the concrete legibility failures are good, but they are close to recent Person Q writing cards and to your own audience-relative checks request." }, { "meeting_id": "SPM16-03", "substance": "Retention audit for the lawyers: no single deletion horizon; Engineer A listed vendor gaps. Person B settled on 90 days, not 14, while the privacy policy is reworked.", "selection_reason": "Skipped: an internal compliance workflow, now resolved. It isn’t a reading-lane moment and doesn’t call for legal commentary." }, { "meeting_id": "SPM16-04", "substance": "Short API/dashboard outage. Person D suspected a ~30k-member install; Person J said his env-var restart likely caused it. Person D said the onboarding scan should survive restarts.", "selection_reason": "Not carded alone. Mentioned only as a possible, unconfirmed link in the UC Davis card." }, { "meeting_id": "SPM16-05", "substance": "Slack agent mode shipped (native Stop, suggested prompts, a continuous panel). Stop appears only in DMs; in channels you type “stop.”", "selection_reason": "Skipped: you already reacted to the launch, and this is a minor discoverability note." }, { "meeting_id": "SPM16-06", "substance": "Person S built your Checks bench. Person T asked about ownership transfer, you worked through access and a render bug, and Person T opened a fix PR.", "selection_reason": "Skipped: your own exchange; nothing new beyond it." }, { "meeting_id": "SPM16-07", "substance": "Person T’s eval definition (“replaying past attempts…”), your SAT versus reference-check framing, Person U’s 30/70 quip, and your audience-relative checks request, which Person Q wants too.", "selection_reason": "Skipped: you took part, and the best framing is largely yours." }, { "meeting_id": "SPM16-08", "substance": "Sales tax for the studio apps is moving to Kintsugi. Person X corrected himself: only main Every’s history has been imported, and the app histories are still pending.", "selection_reason": "Skipped: routine finance and ops." }, { "meeting_id": "SPM16-09", "substance": "All Access sneak peek on Sept 25 plus an Oct 9 camp. Person I wants himself, Person B and Person D there so top-tier subscribers talk to the makers.", "selection_reason": "Skipped: a prior card already covers the demo idea; the new detail is minor." }, { "meeting_id": "SPM16-10", "substance": "Person N and Person H chose sharp corners for cards and buttons and want a place to test designs in context.", "selection_reason": "Skipped: a routine design choice." }, { "meeting_id": "SPM16-11", "substance": "The Rebuilding Compound Engineering story likely slips; the one-week-or-two question is unanswered.", "selection_reason": "Skipped: scheduling." }, { "meeting_id": "SPM16-12", "substance": "You noticed Quasar setting goals for simple tasks. Person F finds it confusing that pausing a goal doesn’t stop the task.", "selection_reason": "Skipped: your own thread, and small." }, { "meeting_id": "SPM16-13", "substance": "Thesis sponsor opt-in is live for the afterparty; Person I is waiting on sponsor names.", "selection_reason": "Skipped: logistics." }, { "meeting_id": "SPM16-14", "substance": "Person T says an a16z contact wants to test the Agent and asks how beta enrollment works.", "selection_reason": "Skipped: secondhand interest with no reaction or outcome." }, { "meeting_id": "SPM16-15", "substance": "Person D’s response to the Vercel-style department-agent idea: by default, find people by Slack job title, with an optional installer choice.", "selection_reason": "Skipped: an incremental design response, not shipped." }, { "meeting_id": "GPM16-01", "substance": "Checks: Person S closed the codebase import of your benchmark and merged a flow where a new account’s AI creates its benchmark over MCP (“There is no form here.”).", "selection_reason": "Skipped: an interesting product stance, but it overlaps with your own Checks thread today." }, { "meeting_id": "GPM16-02", "substance": "Person D’s open answer-only reply PR, approved by Engineer A: five descriptive prompt edits left the footer unused in 0 of 15 runs, while one worked example got it used 5/5. The eval harness also wasn’t sending the per-turn note.", "selection_reason": "Strong runner-up (“examples bind where descriptions do not”). Held back to avoid two prompt-engineering cards; the GPM16-03 priming finding was more counterintuitive." }, { "meeting_id": "GPM16-03", "substance": "Engineer A’s open PR adds retry-once for degraded jobs and plain-language error handling. It found that naming counts as withholdable made the agent drop numbers elsewhere, and that a disclaimer made it worse.", "selection_reason": "Selected as worth_spreading, with the small-sample and open-PR caveats kept." }, { "meeting_id": "GPM16-04", "substance": "Person P’s merged baby-agent PR changes external-agent profiles to eight fields and keeps “Observed behavior” separate, because an agent’s self-description is “a claim, not evidence.”", "selection_reason": "Skipped narrowly: a crisp idea, but experimental and similar to the prompt-lesson card." }, { "meeting_id": "GPM16-05", "substance": "Person P’s merged baby-agent PR adds external AI employees: other bots on a roster that the agent interviews, briefs, and waits on in the thread.", "selection_reason": "Skipped: an interesting agent-to-agent primitive, but not yet run in production." } ], "notes": "All 25 units read in full. The Notion transcripts have no speaker labels, so speakers are local inferences marked 'likely'; the Slack and GitHub authors are verified accounts. GitHub measurements are reported in PR bodies (partly AI-generated) and small-sample. The Person M/outage link is flagged as unconfirmed. Four flags; runners-up were NPM16-03 skill discoverability and GPM16-02 examples-over-descriptions." }