{ "flags": [ { "id": "r1-skills-nobody-cares", "meeting_id": "NPM16-04", "kind": "off_strategy", "title": "Person B: “I don't think anyone cares about skills. Outside of maybe us”", "face": "In Wednesday’s Agent product review, likely Person B called it “maybe a hot take” that nobody outside Every cares about skills, citing usage data. Likely Person D agreed, then set a failure test: if skill updates raise skill usage but satisfaction, revenue and memberships “Don't change. then that's a failure.” He kept skills anyway as the vehicle for automations, asking: “if we had no skills in the product right now, would we ship skills?”", "context": "Every Agent Product Review, September 16: likely Person D was walking through the roadmap (automations becoming “skills on a schedule”, proactive skill creation) when likely Person B raised a worry that skill-centred work might be vanity work that doesn’t move activation. Unlabelled two-person transcript; attribution is from role and turn continuity.", "moment": [ { "line": 99, "speaker": "likely Person B", "quote": "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?" } ], "why": "Your H2 compounding loop is built around the skill as its unit, and the two people running the Agent are openly testing whether users value that unit at all; Person D’s failure test is a ready-made yardstick for the skills roadmap. This was a discussion, not a decision to drop skills, and Person B’s “strong usage data” was not shown.", "strategy_quote": "Improve the skill against historical examples until it performs well.", "reaction_inferred": null, "options": [], "people": [ "Person B", "Person D" ], "confidence": 0.72 }, { "id": "r1-q-reader-context", "meeting_id": "SPM16-02", "kind": "worth_spreading", "title": "Person Q: the model left out the context readers need—four unexplained names", "face": "Person Q posted an email draft as an example of a problem she keeps seeing in model writing: it “doesn't have great judgment on the context a reader needs.” Her list: “Claudie, Andy, Viktor, Fin — four names with no hint they're agents, systems, or people”; benchmark percentages with no baseline; and the essay’s payoff, “Climbed a frame, not the framer,” compressed into “a phrase only the author can decode.”", "context": "Slack, September 16, while Person Q spot-checked early-access models on writing tasks. Separately that day she said intros left her to “reconstruct how the ideas connect,” and that social-idea pitches stayed “navel-gazey” even with her Reader A lens skill. The screenshot of the draft was not inspected.", "moment": [ { "line": 74, "speaker": "Person Q", "quote": "It doesn't have great judgment on the context a reader needs to make information legible to them." }, { "line": 76, "speaker": "Person Q", "quote": "Claudie, Andy, Viktor, Fin — four names with no hint they're agents, systems, or people." }, { "line": 78, "speaker": "Person Q", "quote": "\"Climbed a frame, not the framer\" — the essay's whole payoff, compressed into a phrase only the author can decode." } ], "why": "These are concrete, reusable examples of a failure that is about the reader, not grammar. They fit your request that afternoon for checks judged against an audience brief, which Person Q said she wanted too (my connection; she did not link the two).", "strategy_quote": null, "reaction_inferred": null, "options": [], "people": [ "Person Q" ], "confidence": 0.7 }, { "id": "r1-pitch-vs-findability", "meeting_id": "NPM16-03", "kind": "room_texture", "title": "Person E pitched the Agent as making work visible; Person L couldn’t find a skill", "face": "Planning the September 25 All Access office hours, Person E gave Person L the pitch: with one company agent in Slack, “your whole team gets insight” into how others use it. His example was Person I’s Remotion video skill, which he now uses constantly. Likely Person L’s reply: he’d tried to find that skill, but “Just asking the agent didn't lead to the right things,” so he started using Remotion himself.", "context": "Person L x Person E, September 16: Person L runs All Access office hours and wanted Person E to present the Agent before launch. Person E recounted the ‘we were wrong about OpenClaw’ story and the pitch, then talked about the video skill Person I built. Unlabelled two-person transcript; attribution is by content (Person L builds lesson demo videos).", "moment": [ { "line": 120, "speaker": "likely Person E", "quote": "with the one company agent in Slack, your whole team gets insight into like watching people do other stuff with the agent and it kind of AI pills the whole organization. That's like our pitch." }, { "line": 145, "speaker": "likely Person E", "quote": "he is the one that came up with the re-motion skill that the agent has." }, { "line": 153, "speaker": "likely Person L", "quote": "I started playing with Remotion myself because I, to be honest, didn't know how to find the skill. Just asking the agent didn't lead to the right things or whatever." } ], "why": "Seconds after hearing the show-don’t-tell pitch, a colleague gave a first-hand miss on the same promise, inside Every. It’s internal, so it isn’t a message test, but it’s a concrete gap to keep in mind before the office hours. The miss is Person L’s account; no agent logs were checked.", "strategy_quote": null, "reaction_inferred": null, "options": [], "people": [ "Person E", "Person L", "Person I" ], "confidence": 0.62 }, { "id": "r1-priming-disclaimers", "meeting_id": "GPM16-03", "kind": "worth_spreading", "title": "Telling the Agent to hide job counts made it drop timings from unrelated replies", "face": "Engineer A’s PR tells the Agent to stop repeating a provider’s internal debugging details, like “112 candidates and kept 5”. In testing, the new rule made the model drop real figures (250ms, 8s) from replies where nothing had failed. Adding “never overrides the numbers rule” did worst, at 0/3. The PR’s conclusion: “Priming does not read disclaimers.” The fix deletes every mention of counts or numbers.", "context": "EveryInc/plus-one #1755, open; head commit September 16, 19:05 UTC. A copy-edit job returned success with a crashed stage inside; the Agent relayed the jargon and didn’t retry, and the user said “i dont know what you're talking about.” The PR body appears agent-drafted (Claude Code footer); pass rates are author-reported on 3-run samples, and the final variant was still pending when written.", "moment": [ { "line": 70, "speaker": "Engineer A (PR description)", "quote": "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 description)", "quote": "This scored **worst (0/3)**. Priming does not read disclaimers; a reassurance names the class as loudly as a ban." }, { "line": 78, "speaker": "Engineer A (PR description)", "quote": "Every mention of counts, tallies, candidates, attempts, figures and numbers is now deleted from the addition, in both directions." } ], "why": "This is a counterintuitive prompt-writing lesson that applies to anyone writing Agent or Codex instructions: telling a model “don’t say X” can suppress everything that resembles X. The samples are small and it’s unmerged, so read it as a strong anecdote, not a settled result.", "strategy_quote": null, "reaction_inferred": null, "options": [], "people": [ "Engineer A" ], "confidence": 0.55 } ], "reading_notes": [ { "meeting_id": "NPM16-01", "substance": "Agent launch standup reviewing Person H’s landing-page design: skill section, a permissions example, pricing (drop enterprise, ‘free to start’), and likely Person I saying the compounding diagram ‘makes me feel like I have to do all of these things’, which led to ‘never repeat yourself’/‘never make the same mistake twice’ header ideas.", "selection_reason": "Not selected: prior cards already cover the compounding-messaging shift; the header idea was a close runner-up." }, { "meeting_id": "NPM16-02", "substance": "Onboarding for Person K, a ghostwriter whose Slack is mostly clients; she asked whether the Agent is paid, wanted a Claude-projects connection, set up a call-prep automation, and said her Claude Cowork automations ‘just didn’t fully work’.", "selection_reason": "Not selected: real reaction, but the Agent wasn’t pitched with the ambassador message, and the outcome of the automation is unknown." }, { "meeting_id": "NPM16-03", "substance": "Person L and Person E plan the Sept 25 All Access office hours around the Agent; Person E gives the visibility pitch, and Person L says he couldn’t find Person I’s Remotion skill by asking the agent; both describe building tools instead of repeating agent tasks.", "selection_reason": "Selected (r1-pitch-vs-findability): a concrete, first-hand gap against the pitch, right after the pitch." }, { "meeting_id": "NPM16-04", "substance": "Agent product review: activation roadmap (<20% of workspaces accept an automation; two of three see a suggestion and decline), automations becoming skills on a schedule, Person B’s ‘nobody cares about skills’ challenge, spec/kickoff process, billing friction and retros.", "selection_reason": "Selected (r1-skills-nobody-cares): a strategy-relevant tension with a concrete test; roadmap/process detail left out." }, { "meeting_id": "NPM16-05", "substance": "Onboarding for Person M, IT executive director at UC Davis Engineering, who installed into a ~30,000-person Slack, wanted private-channel-first testing, and drew a line between internal data and legally protected student/defense data.", "selection_reason": "Not selected: a vivid external reaction, but not a message test and no outcome yet; a strong runner-up." }, { "meeting_id": "SPM16-01", "substance": "Person N’s one-prompt golf and sourdough games on an early-access model (pelican that steals the ball, ‘rent $4200’); an external collaborator said ‘This is WILD’, and Person P called it ‘astra slop, with taste’.", "selection_reason": "Not selected: fun model-testing texture, but thin and about an unverified early-access model." }, { "meeting_id": "SPM16-02", "substance": "Person Q’s writing spot-checks on early-access models: coherence problems, generic social closers, social pitches ‘navel-gazey’ even with her Reader A lens skill, and an email draft that lacked the context readers need.", "selection_reason": "Selected (r1-q-reader-context): specific, quotable writing-judgment examples." }, { "meeting_id": "SPM16-03", "substance": "Retention audit for lawyers: Person R and Engineer A list per-vendor retention gaps (backups, Browserbase profiles, subprocessor list); Person B settles deletion at 90 days and says the privacy policy is being reworked.", "selection_reason": "Not selected: security/legal lane with owners in place and decision made; no reading-card value." }, { "meeting_id": "SPM16-04", "substance": "Brief Agent dashboard/API outage; Person D first blamed the ~30k-member install, then Person J said an env-var change restarted the server; Person D says onboarding should survive restarts.", "selection_reason": "Not selected: routine incident, cause corrected; don’t link it to the UC Davis install." }, { "meeting_id": "SPM16-05", "substance": "Person J announces Slack agent mode (native stop, suggested prompts, continuous transcript); Person I and Person B couldn’t find Stop outside DMs; typing ‘stop’ works in channels.", "selection_reason": "Not selected: you reacted to the launch; discoverability nit is minor." }, { "meeting_id": "SPM16-06", "substance": "Person S created your Checks bench from your task dump; Person T asked how to hand ownership over; you found it, asked about ratings, and reported a render bug that Person T is fixing.", "selection_reason": "Not selected: your own exchange; routine setup." }, { "meeting_id": "SPM16-07", "substance": "Person T frames evals as ‘replaying past attempts’; after your SAT-vs-reference-check line he adds red/green flags that inform judgment without being counted; Person U: references are ‘70% were they good to work with’; you ask for audience-relative checks, Person Q agrees.", "selection_reason": "Not selected: you were in it; used as supporting context for the Person Q card." }, { "meeting_id": "SPM16-08", "substance": "Sales-tax integration for Studio apps via Kintsugi; Person X corrects that only main Every history is imported so far, and apps should not backfill.", "selection_reason": "Not selected: operational finance detail." }, { "meeting_id": "SPM16-09", "substance": "Office hours/camp planning after the launch moved to Oct 6; Person I wants himself, Person B and Person D in the room because All Access is ‘our highest tier’.", "selection_reason": "Not selected: logistics; the sneak-peek idea is already in prior cards." }, { "meeting_id": "SPM16-10", "substance": "Person N and Person H settle on flat, sharp-cornered buttons and cards for the Agent dashboard after Person H asked to see changes in context.", "selection_reason": "Not selected: routine design call." }, { "meeting_id": "SPM16-11", "substance": "Person Q says the Rebuilding Compound Engineering story needs more time because of Vibe Check load; one- or two-week delay unanswered.", "selection_reason": "Not selected: scheduling." }, { "meeting_id": "SPM16-12", "substance": "You noticed an early-access model triggering ‘goals’ on simple tasks; Person F finds pausing a goal while agents keep working confusing.", "selection_reason": "Not selected: your own thread; minor UX." }, { "meeting_id": "SPM16-13", "substance": "Person I says sponsor opt-in is live for the afterparty and he’s waiting on sponsor names for the rest.", "selection_reason": "Not selected: logistics." }, { "meeting_id": "SPM16-14", "substance": "Person T says a16z’s Person AG wants to test the Agent and asks how to enroll him.", "selection_reason": "Not selected: single unanswered request; no reaction to the message." }, { "meeting_id": "SPM16-15", "substance": "Person I passes along a Vercel idea for department subagents; Person D says onboarding could find department people from Slack job titles, with an optional installer pick.", "selection_reason": "Not selected: early idea with a brief response; no decision." }, { "meeting_id": "GPM16-01", "substance": "Person S closed Person T’s PR that imported your Checks benchmark via migration (benchmarks go through MCP CRUD instead) and merged a PR letting a new account create a benchmark through its own AI.", "selection_reason": "Not selected: routine architecture choice." }, { "meeting_id": "GPM16-02", "substance": "Person D’s open PR making Agent replies answer-only, with caveats in a grey footer; five rule descriptions left replies unchanged, and a worked example moved footer use from 0/15 to 5/5; the eval harness had been omitting the per-turn note. Approved by Engineer A this afternoon.", "selection_reason": "Not selected, narrowly: strong prompt lesson (‘examples bind where descriptions do not’), but the experiment is from Sept 14–15 and I kept one prompt-lesson card, the more surprising one." }, { "meeting_id": "GPM16-03", "substance": "Engineer A’s PR for a job that ‘succeeded’ with a crashed stage: retry once and translate provider jargon; banning count words primed the model to drop real figures everywhere, and a disclaimer made it worse.", "selection_reason": "Selected (r1-priming-disclaimers)." }, { "meeting_id": "GPM16-04", "substance": "Person P’s baby-agent PR aligns ‘external AI employee’ profiles to his skill, with eight fields and the rule ‘an agent's self-description is a claim, not evidence’; two agents never answered the interview.", "selection_reason": "Not selected: nice framing, but mostly schema work; runner-up." }, { "meeting_id": "GPM16-05", "substance": "Person P’s baby-agent can put other Slack bots on its roster and hand them a brief, then resume the same turn when they reply; ‘never mention another bot except to brief it’.", "selection_reason": "Not selected: interesting multi-agent plumbing, but engineering detail with no user outcome yet." } ], "notes": "Notion transcripts are unlabelled (speaker attribution by direct address, role and turn continuity; ‘likely’ used). Slack and GitHub have author labels; the GitHub PR bodies appear agent-drafted under the author’s account. No generated summaries used. Runners-up: NPM16-05 (UC Davis IT director installing into a ~30k Slack), GPM16-02 (examples beat descriptions), GPM16-04 (agent self-description is a claim, not evidence), NPM16-01 (‘never make the same mistake twice’ header). No message_tested moment qualified: nobody pitched the ambassador message to an outsider and quoted a reaction." }