Day 1 — Grok Bot Galaxy full builder notes (Code Solutions)

Day 1 — Grok Bot Galaxy (full timed notes)

Source: https://x.com/bot/status/2099890276315848743 (replay ~8:45:19) Hosts: Matt Palmer (@mattyp), Lauren Tan (@poteto), Roshan Sadanani (@roshan_s) Flag: Metrics/claims as stated on stream unless noted — not independently verified. Contest/prize fine print [unverified].

Builder pack consolidated from live scribe drops (player-time order). Prefer these timed bullets over inventing detail. BRB/empty stretches marked; gaps labeled when no notes were captured.

Arc (short)

1. Open — Build a company in 3 days with Grok Bot; idea TBD; Starship-launch contest (follow @grok/@bot, quote + template; deadline Sept 29 as said) [unverified fine print].
2. Grok Bot 101 — Roman + Amrita: bots as teammates with own computers + cloud; demos (Data Dan / Slide Sonya / Email Ethan); teach-a-task → skills; Auto-review; Marketplace; agent-to-agent; Manager routines.
3. Build kickoff — Research bots (Marky McMarkface); lean pop-up OS; company Potato Lab / org shipbythursday; domain shipbythurs.day; pStack, Tater/Hashbrown; Vercel + waitlist (Resend/PlanetScale path).
4. Guests — Peter Yang; Codie (distribution, Proof Vault); Lingxi (eng fleet / auto-fix CI); Kevin Niparko + Roshan (PM demo); Eric/Karat; Shub (founder bots); Jenny (~22 bots, Scout venue advice).
5. Product pivots on stream — Restaurant pop-up → art exhibition → merch + ticketing + venue pop-up (dogfood own event first). Target date floated: Oct 15 (as said on stream).
6. Close — Infra stood up (Slack/Notion/GitHub/Vercel/V1 lander); contest plug; end ~524:50.

Timed notes

~0:00–15:00

Opening (~0:00–10:00)

• Setup: three people at a table under a “Grok Bot Galaxy” backdrop; replay length 8:45:19 (~1.4M views on the player) [unverified view count].
• Post names Matt Palmer (@mattyp), Lauren Tan (@poteto), Roshan Sadanani (@roshan_s). No on-screen lower-thirds early; speaker IDs from faces alone not locked at open.
• Framing: Day 1 of a three-day Grok Bot build (research → plan → product; later sessions mentioned on the post).

~0:00–15:00 (premise + contest)

• Premise: Matt, Lauren, Roshan will build a company in 3 days live with Grok Bot / AI tools. Idea not chosen yet.
• Roles (as stated): Matt — Developer Experience (mentions Lee Rob, Eric, Milosh); Lauren — Grok Bot at SpaceX AI; Roshan — Product (connectors / Grok Bot marketplace).
• Contest: “Grok Bot Galaxy livestream challenge” — winner + guest trip to a Starship launch; runner-up tour of SpaceX Rocket Factory (Hawthorne). Entry: follow @grok and @bot, quote the official post with bot description + share-template link. Deadline Sept 29. [unverified beyond what was said on stream]
• Upcoming named: “Grok Bot 101” with Roman and Amrita.
• Use cases: Matt describes a bot that watches X bookmarks for npm packages → Cursor cloud agent → Cloudflare preview deploy. Lauren: automates X replies to “potato potato potato”; performance / contrib setup.
• Next step they name: market research from “thousands” of user suggestions; brief “what is a business?” (useful, dogfood, make money).
• No on-screen demos yet in this stretch.

~15:00–30:00

• Still no product pick. Focus on something people will pay for; Lauren notes bots can get a Stripe “link card” / credit card in Grok Bot. Roshan floats bots creating accounts and buying from the business.
• Stack talk: Slack, Notion, Linear — Lauren prefers PR-over-issue; mentions wiring “Michelin Kitchen.” Introduces pStack (open-source Marketplace plugin of her workflows/skills). Claim: ~2,500 pStack PRs shipped last month (~83/day joked); “merge to main” / commit-meter bit. [unverified metrics]
• Clean slate: empty GitHub org github.com/shipbythursday; new Grok Bot accounts with zero bots. Plan: install pStack, user-research bots from templates, “founding engineer” + “intern” bots. Matt: bots can create bots; handbook/onboarding bot idea.
• Contest recap (Starbase Starship trip / Hawthorne runner-up). Shoutout to ThePrimeagen. Claim ~80k viewers (live at the time) [unverified]. App download plug; long continuous hours mentioned.
• ~29:28 handoff to Grok Bot 101 with Roman and Amrita → BRB screen. Still no on-screen demos this stretch.

~30:00–45:00 (Grok Bot 101 — Roman + Amrita)

• Roman + Amrita take the stage after BRB. Roman’s AI maturity curve: Ask (chatbots) → Do a task (copilots) → Delegate outcome (bot) → Staff function (team of bots).
• Grok Bot design pillars (Roman): teammate paradigm (role bots with memory), computer access (not only APIs/MCPs — own desktop), cloud execution (always-on, not laptop-bound). Value props: easy as iMessage, always-on, uses your tools, finishes work, shareable workflows.
• Amrita demo team: Data Dan, Slide Sonya, Email Ethan — collaborate without the user micromanaging (form → slides/charts → email).
• Live app: creates Data Dan; voice-dictates a Google Form (coffee cups/day + SF shops: Ritual, Philz, Sightglass, Blue Bottle). Notes conversational voice back-and-forth shipping internally “within the next week or so” (as said on stream).
• Shows bot’s Linux VM desktop; Marketplace plugins (GCal, Drive, Slack, GitHub, Jira, Notion). Approval prompt “Bot wants to run a task” → Allow once (enterprise guardrails).
• Teach-a-task: Amrita drives Slide Sonya’s cursor in Google Slides (Fly-in animation). Segment ends mid-demo.

~45:00–60:00 (Amrita demo continues)

• Teach-a-task → skill: stops recording; Slide Sonya turns the Fly-in animation into a reusable skill.
• Parallel bots: Data Dan building the coffee Google Form live on his VM (QR code planned for audience); approvals via Allow once.
• Auto-review Rules: granular allow/ask (e.g. ask before email replies; auto for slides). Demo rule blocks Email Ethan until Allow once.
• Architecture: isolated Linux VMs per bot — “One Grok Bot cannot access another Grok Bot’s computer.” Human + bot can co-edit same deck without conflict.
• Routines: e.g. Slide Sonya daily 9am slide-change summary; multi-bot routines possible.
• Agent-to-agent: Email Ethan messages Data Dan / Slide Sonya on its own for context (“Amrita asked me…”). Orchestration framed as building your own team/company of agents.
• Persona: Email Ethan learns tone from prior emails; long-lived memory across projects.
• Claim: computer use improvements “over the next couple weeks” (Grok Bot / Cursor / SpaceX generally — as stated). Clip ends mid-sentence ~60:00.

~60:00–75:00 (Amrita wraps 101)

• Duplicate bots / share as template for company-standard roles (e.g. brand design). Marketplace: featured public bots (names cited: Lauren Tan, Lenny Rachitsky, Claire Vo) + team plugins.
• Email draft modal: edit recipients/text, rephrase, without living in the mail client.
• Data Dan → QR for coffee form; audience reports “no access” → Amrita steers fix (“make link public”); bot Auto-review opens Forms share settings. Feedback → memory (she says S3).
• Folders + group chat (Ethan + Dan + Sonya). Group prompt: chart coffee data on last slide, then Ethan emails Jason — Ethan holds until Sonya’s slide lands.
• New “Manager” orchestrator bot + routine every 2 hours for team updates/blockers.
• ~74:33 Amrita hands back to Matt / Roshan / Lauren; leaves bots running on screen.

~75:00–90:00 (Amrita Q&A)

• Auto-review: classifier + custom rules (e.g. ask before mkdir in terminal). Trust builds via memory of prior approvals.
• Multiplayer humans+bots in Grok Bot group chat: “coming very soon”; Slack bot tagging already a use case.
• Enterprise: credentials / VPN for internal tools; “just released” strong 1Password integration — secrets via in-app modal, not exposed to SpaceX side (as claimed) [unverified architecture claim].
• Teach-a-task skill example: Add Google Slides object animation. Marketplace callouts: Claire Vo Traitbot, Lauren Tan AI eggplant, Eric Zakariasson Projects Manager, Lenny Rachitsky Overheard.
• Contest slide again (quote post + bot description + template link + follow @Grok & @Bot). Memory in S3; editable/forgettable; duplicate bot = persona copy, fresh empty memory.
• Takeaway slide: computers without MCP-only limits; memory + video teaching; human approval for enterprise actions. Advice: automate the most annoying daily task.
• ~89:38 Amrita ends → BRB.

~90:00–105:00 (build + Peter Yang)

• BRB until ~91:03. Panel: Matt, Lauren, Roshan + guest Peter Yang (ex-PM; YouTube / “Behind the Craft”). Peter: runs a one-person business on Grok Bot — “talk to my bots all day instead of humans.”
• Schedule named (as said, PT): Eng 12:30–2:00 · PMs w/ Kevin 2:30–3:30 · Founders w/ Shub 4:00–5:30. (That’s ~2:30–4 / 4:30–5:30 / 6–7:30 CT.)
• Foundations: GitHub shipbythursday, Slack, Notion — Lauren pushes scrappy/fast over heavy docs.
• First research bot: Marky McMarkface (chat-suggested name). Roshan screen-shares: X marketplace plugin connected; prompts synthesis of replies to their “build a business” posts.
• Bot output themes (as scrolled): not another SaaS demo; hardware; “company that builds companies”; pitches like Neopets/Club Penguin–style consumer. Lauren: aim for mom-level accessibility. Side bit: “Grok Pot” rice-cooker joke tweet.
• Still no final product pick. Screen-share AV hiccup ~2.5 min before share landed.

~105:00–120:00 (research → pop-up OS lean)

• More Marky themes: company-that-builds-companies; use the ~70k audience as customers [unverified]; pitches like Pokémon-card marketplace, indie-author book marketing, M&A advisory. (Peter not on cam this stretch.)
• PM shift: Lauren — skip long PRDs, prototype + prompt bots; status “why not today”; typing feels like “casting spells.”
• Direction: digital ↔ physical / neighborhood. Bot reacts to heart emoji; name ideas “Open Brief,” “Kitchen Desk.”
• Roshan’s “Market Research Master” (ad-hoc X connector): top pick among 5 → Pop-up OS for restaurants (SF limited-run concepts). Also: adopt-a-restaurant, dinner theater, ghost kitchen, concierge. Panel leans #1 (events, ticketing, tables, payments).
• Start prototyping: Lauren creates prototyping bot → imports Marketplace dr eggbot. Matt’s dr eggbot installing pstack; “steve” lists options (LiveSignal, Local shop kit, etc.).

~120:00–135:00 (Potato Lab lock-in)

• Locked on pop-up restaurant platform. Lauren dictates a scrappy plan to “steve” (Google Sheet as starter DB). Matt (off-cam): prospecting bot via dr eggbot to find SF restaurant owners.
• Lauren: dr eggbot rewrites steve → chief of staff / EA, then spins a prototyper bot named grokpot (chat-suggested). Matt creates private GitHub repo pop-up.
• Company name from chat: Potato Lab. Matt adds a Prioritizer bot; Lauren has grokpot pull landing-page refs / mood board before code.
• Dogfood idea: run a real physical pop-up (e.g. fries downstairs) → need ≥2 landing pages (platform for restaurateurs + consumer page for the event).

~135:00–150:00 (lander themes + tater)

• Roshan’s steve (CoS): voice dump → “restate what I said…”; Potato Lab framed as operator platform + dogfood pop-up + attendee surfaces.
• grokpot Lofi waitlist directions: linear-dark / notion-minimal / arc-want (HTML/CSS preview without Cursor yet). Round 2: Night-market (“Spin up a stall. Own the night.”), Ticket-stub, City-pin.
• Peter Yang: enjoy-what-you-build; pure SaaS landing+DB hard to charge for; wants more bot proactivity (e.g. weekly metric crons). Then leaves.
• New engineer bot tater (via dr eggbot). Stack suggestion: Vercel, PlanetScale, Next.js, Tailwind, shadcn, Postgres, Drizzle, Clerk, Stripe (friends from Vercel/PlanetScale said to join later).
• Notion company doc: placeholder name “Ship by Thursday”; Lauren CTO / “Chief Potato Officer”; Roshan CPO; CEO unresolved on cam.

~150:00–165:00 (Vercel lander ship)

• Theme: Roshan picks pink for PLATFORM waitlist (via grokpot in Slack). Lauren sets Slack title “chief potato officer”; prompts @steve in #all-shipbythursday to build landing page → shipbythursday/popup.git → Vercel.
• ~3 min later: grokpot pushed index.html; Vercel preview up. Lauren added as Vercel org owner; authorizes GitHub. Deploy on main confirmed after AV glitch (~158 BRB).
• Marketing bot domain ideas (claimastall.com, nightmarket.app, pinapopup.com…). Considering shipbythursday.com (chat may have sniped — unverified).
• Matt spinning Ops + Creative Director bots. Signup storage: Notion MCP first → pivot to PlanetScale Postgres (“small tweak… use planetscale postgres instead”).
• Role note on cam: Roshan referred to as CEO/CPO in this stretch (earlier CPO; CEO was fuzzy).

~165:00–180:00 (domain + Codie arrives)

• Guest no-show → keep building. Lauren: steve tells tater to use cloud agents / project agent; Grok Bot orchestrates, Cursor harness codes; work locally on cloud VMs while PlanetScale creds pending.
• Domain acquired: shipbythurs.day (company-behind-the-companies lander; product gets its own domain later). Matt: dr eggbot + Cursor team kit → cloud agents for lander.
• Lauren: steve opened a ~2k-line PR → new rule: no PRs, ship to main until someone yells; wants a code-reviewer bot. ~172 BRB for AV.
• Guest Codie (video): ~15M followers claim [unverified]; Vercel book-launch dashboard (~135k users; CAC/AOV) [unverified]; PULSE. Advises sell to 3 people before ops/legal; cites ~430k live viewers (as said) [unverified]. Shows controversial-tweet distribution + Higgsfield AI video. Matt restates pop-up → ops → sellable software.

~180:00–195:00 (Codie — marketing / ops)

• Marketing: controversy OK; don’t run paid ads until ~5–10M organic traction (her bar) [unverified]; “steal the frame, not the verbiage”; imitate → iterate → innovate. Claims: 2.4M views on one video; book launch 5.7M views / ~2k signups; Contrarian Thinking Capital ~36 investments [unverified].
• Ops: don’t “hire A-players and disappear” — track cash-on-cash, AOV, LTV. Distribution is a moat; #1 AI-startup failure mode = no distribution (as claimed). Build a Proof Vault of testimonials/claims from day one. Lo-fi creatives beat polished corporate ads (~40% relative, her figure) [unverified].
• Live demo: hosts spawn Lead Capture + Operator Research bots; research SF restaurateurs; Notion founding docs shown. Codie on traditional/home-service businesses + AI as clip ends.

~195:00–200:00 (Codie wraps)

• SMB AI: most still light ChatGPT use. Sell cash outcomes; ~15% margins → hard to sell AI not tied to immediate revenue [unverified].
• Lead magnets: her example 74-question archetype survey (Claude/Vercel/Replit) on ~15k B2B owners [unverified]; give 10x value vs asking for email. Slide: call out A-players in public.
• Parting: profit is the point; “if you build it, they will not come.” Roshan recaps revenue-first + tech as enabler.
• ~199: Matt: short break, then build mode + landing page. BRB.

~200:00–225:00

• gap: no notes captured (adjacent drops: Codie wrap → BRB ~199; next notes resume BRB until ~228:44). Clips on disk for 195–210 and 210–225; treat as break / empty stretch unless re-watched later.

~225:00–240:00 (build resume + Lingxi starts)

• BRB until ~228:44. Recap: aligned on SF pop-up restaurant OS/software for the 3-day build (research + Peter + Codie).
• Lauren “potato factory”: new Hashbrown bot reviews PRs from Tater; adding pstack via Cursor plugins; Cursor automations so PR → Slack → review → maybe merge. Goal: “survive until the next day.” Host: Grok Bot to ideate → Cursor Cloud Agents.
• ~234: Grok Bot for Engineers — Lingxi Li (SpaceX AI / ex-Cursor; built Cursor 3 agent window). Claims: uses only Grok Bot for eng last ~2 months; built first Grok Bot mobile alone in ~3 weeks; previously ~15 cloud agents [unverified].
• Maturity curve: Tab → Ask/Edit → Cursor 3 agentic → Cloud Agent automations (24/7, 10+ parallel, “10x”) → Grok Bot autonomous orchestration. Grok Bot: autonomous agents, MCP tools, manage Cloud Agents, memory & routines. Clip ends mid-feature pitch.

~240:00–255:00 (Lingxi — pillars)

• Grok Bot pillars: 24/7 agents; MCP tools (Notion/Figma/Slack…); first-party Cursor Cloud Agent management (read transcripts, start private workers, judge completion/proofs); long-horizon named-bot memory & routines (learns corrections → future agent prompts).
• Why: own computer (no “caffeinated laptop”); multi-platform — recently shipped iPad + Android; also iOS/Windows/Linux/macOS (as claimed). Remote control when login/captcha blocks the bot.
• Away use cases: Slack PR reviews while asleep/flying; verify screenshots/tests then cloud-agent scan; catch X bugs via Slack MCP → repro on main → fix PR.
• Human role: product/design/architecture/hard perf; teach boundaries (“say no”) into memory.
• Nightly code cleanup (e.g. 3am) with auto-merge if clear proof. TestFlight seats via Slack mention (one-prompt internal tooling). Starts “Auto-fix Everything” (CI/deploy) as clip ends.

~255:00–270:00 (Lingxi demo — auto-fix / fleet)

• Auto-fix CI: bot spots red build → cloud agent fix → merge; claim “most of the time… within 10 minutes.” [unverified]
• Fleet model vs one mega-bot: Lingxixi (CoS), Craig (UI), Steve (Devex), Hogan (Infra), Jenny (Ops) — separate context/workflows; humans talk to CoS. (Demo naming drifts: primary bot used as “Craig” for routing.)
• Marketplace install “Nightly Audit Engineer” → rename Steve; Craig onboards via agent-to-agent dump (Notion boards, PR hygiene).
• Live bug: flylo-air.com /trips stub — Craig’s browser confirms. Jenny onboarded for ops playbooks; Steve already spinning nightly-audit cloud agents.
• P0 care: “urgent” makes agents guess; don’t babysit traces — set routine every 5 min to check cloud agents (interrupt sleep/off-goal). Boundaries via normal chat, not only coded skills.

~270:00–285:00 (Lingxi wrap → panel)

• Lingxi wrap: Jenny (ops) broadcasts P0 playbook to fleet; PRs need proofs (UI screenshots / perf metrics). Shared Notion-like FlyLo Engineering Fleet board for multi-agent state. Bug/security bots in “watching.” Cloud-agent PR auto-filled with screenshots + prompts.
• Takeaways: treat bots like interns (chat, not long skill prompts); automate the workflow one level up; start with success/fail feedback loops. Hands back ~279 → brief BRB.
• Panel return: SF pop-up platform (attendees ↔ creators). Roles stated: Matt CEO; Roshan CPO; Lauren CTO / Chief Potato Officer. Matt authorizes Cursor on GitHub for shipbythursday/thursday + company. Operator Research bot lists SF restaurateurs (via dr eggbot factory). Split-screen into Lauren’s workflow as clip ends.

~285:00–300:00 (Lauren / Matt build workflows)

• Lauren: #pr-reviews Slack + Cursor/pstack auto-review (correctness, risks, missing tests). Plans #bug-reports — X feedback → channel → bots open PRs/triage. Explains create-verification-skill (deterministic CLI + feature map) so bots don’t keep asking humans to click around; steve → tater.
• Matt: polishing lander in shipbythursday/thursday with Founding Eng bot — more minimalist. Chooses Resend over Loops for waitlist email. Form already wired to DB (Lauren surprised). Domain lander live.
• Lauren: redesign signup UI — guest waitlist primary, restaurateur help CTA top-right. Brainstorm with tater/cloud agent: login, events, invites, venue map, staff, RBAC.
• Matt MVP: scrappy prototype first; hero flow = group runs a pop-up (venue, staff, menu, plan, scheduling).

~300:00–315:00 (Oct 15 + venue research)

• Min scope: event name/date/time + attend/schedule. Lauren pushes dogfood — run their own pop-up while building. Matt: emails collecting; need SF restaurant research/canvass; lander doesn’t advertise event details yet.
• Target date named: October 15 (to ground comms). Priorities: people → date → location. Google Drive partner list planned.
• Lauren: PR #10 “Add end-to-end verification skill for PLATFORM waitlist”; dictating to steve → tater venue-finder prototype. Roshan posts DB schema in #all-shipbythursday.
• Matt: Bland MCP for outbound calls; ~15 kitchen/restaurant productions listed via bot. Prefers “bot approach” (find 20 businesses + emails/phones) over heavy software first.
• Maps: OpenStreetMap instead of waiting on Google Maps API. Auth: keep scrappy — Vercel deployment protection / Clerk / WorkOS discussed.

~315:00–330:00 (auth / flyers / architecture)

• Auth: Vercel team deployment protection for venue-finder. Roshan: knowledge-base manager bot (updates Notion only when asked); guerrilla flyer idea — top 50 SF street corners via bot.
• Audience recap (~319): SF food pop-up, Oct 15 target; bot teams on design/lander/architecture. Widget peaks cited ~26 bots / 547 messages (on-stream overlay) [unverified].
• Architecture: codebase as agent memory; stay TypeScript (“agents love TypeScript”) vs joke Rust rewrite. Flyer emotion headline brainstorm (“When did you last try something new?”). Merch / exclusivity for event.
• Verification skill: always-on; PRs with visual impact must include video/screenshots. Resend domain verification for email. Glimpse of lander copy “SPIN UP A STALL. OWN THE NIGHT.” PRs noted: #12 waitlist verification, #14 OSM venue finder.

~330:00–345:00 (pivot: restaurant → art exhibition)

• Venue finder still vanilla HTML/JS prototype — ship first, refactor later. Matt presents Grok Bot for PMs at 2:30 (his clock); leaves ~342 to prep.
• Venue bar: ~100 people SF; examples The Pearl / Dogpatch Studios; permits/alcohol lead times are the hard part (e.g. Fort Mason review days). Knowledge-base bot: every 5 min ping active bots → Notion.
• Pivot: restaurant pop-up → Grok Bot–themed art exhibition (fewer food/liquor hurdles). Artists submit digitally; agents review; physical show. Venue finder stays but generic (mixed-use / warehouses / open spaces; restaurants later). Menu bot spun up then superseded by pivot.
• PRs in view: #14 OSM venue finder; #15 waitlist persistence; #16 app-wide verification media requirement.

~345:00–360:00 (art pop-up + PM session starts)

• Build: art pop-up / large warehouse focus confirmed; SF Top 50 Flyer Corners map in #marketing. Engineer + verifier bots busy on PRs. Contest shoutout — Starbase Starship trip for winner + guest (same challenge terms as earlier) [unverified fine print].
• ~353 BRB → Grok Bot for PMs on stage: Kevin Niparko + Roshan (SpaceX AI Product). Framing: building Grok Bot with Grok Bot. DHH quote on software = PM questions.
• “Why Grok Bot” / agents-as-colleagues: not at home in a chat box; smart desk neighbors. Principles started: (1) tie tools for outcomes across Jira/Notion/Figma/Slack, (2) long-running context/memory, (3) independence → own persistent computer, (4) messaging — clip cuts mid-point.

~360:00–375:00 (Grok Bot for PMs)

• Def: “Grok Bot is an agent with a computer. It feels like texting a colleague.” Principles: tools across apps, long-running context, own computer, async messaging.
• PM primitives: Attention List; research across Gong/Zendesk/Databricks etc.; Shipping (decompose → allocate → review → integrate). Claim: Grok Bot = “double-digit percentage of merged PRs” internally [unverified].
• Roster: Cora (CoS), Emily (EM + 5 eng bots), Ashley (data), Pete (product/PRDs), Pixel (design), Rae (recruiter). Why many: referenceability, scoped memory, parallelism.
• Demo: Marketplace plugins; Eng pod standups; Ashley on FlyLo — mobile vs web tickets (demo numbers), charts despite typos, 6am routine; funnel drop at seat selection.

~375:00–390:00 (PM demo wraps)

• Ashley → PM Pete: mobile funnel PRD in Notion. Correction: biggest leak search → fare (~33.9%), not seat selection [demo numbers — unverified]. Pixel: Figma Option A/B; audience picks A. Emily (EM) breaks work → cloud agents → frontend PR.
• VM + “Teach a task” for bespoke workflows. Quote: “These are colleagues with a computer.” Agents often prompt better than humans.
• Lessons slide (“what works past the learning curve”) → thanks. Kevin flags Matt/Lauren/Roshan 3-day Grok Bot startup stream. ~389 BRB.

~390:00–405:00 (Eric / Karat joins)

• BRB until ~401:18. Guest Eric (Karat Financial co-founder/CEO; ex Meta/Instagram PM from 2016). Matt back from PM session.
• Eric on PM: (1) what to build (2) how (3) coordinate — AI makes #3 easy; 1–2 still human.
• Build track: dogfood their own Grok Bot pop-up first as sole user, then generalize (bookings, venues, invitations). Value prop framed around IRL pop-up events (art/warehouse pivot from earlier still in play unless they re-widen).

~405:00–420:00 (with Eric — merch lean)

• Audience lean: first-time pop-up runners who don’t know the business side; IRL focus. Direction workshop: merch pop-up for tech companies — constants date/time/venue/registrations; Matt boils needs to merch + ticketing + venue.
• Matt’s bot roster: Growth Eng, Founding Eng, image generator, Creative Director, CoS. Lauren: prototyper bot drives tldraw in-browser for lo-fi wireframes (steve chat + drawing on cam).
• Merch brainstorm: beyond tees — tungsten cubes/sphere-as-Grok Bot (~$500 joke). Bot templates: share memories/context/instructions without sensitive chat data; passwords stay secure. Scope check: if outputs get flustered/nonsensical, scope is too broad.

~420:00–435:00 (three pillars locked)

• Three pillars locked for the pop-up: merch · venue · ticketing. Art-exhibition ticket wireframe in Excalidraw; collectible/physical tickets (PCB-badge inspiration). Merch: novel over hoodies — tungsten Grok Bot (~$3k joke), plush army, beanie/toboggan bit, programmable desk toys (sales milestone lights, etc.).
• drop bot: Slack poll + Grok Imagine product shots (Desk Torch, Signal Socks, Night Launch Folio, Soft Blob Plush). Roshan: 3D merch gallery in registration; meta line — bots building a company that builds physical bots. Holographic “Platform / Night Market” ticket concept shown.
• Stretch: don’t build creator platform yet — dogfood own pop-up. Roshan rejoins ~426.

~435:00–450:00 (Diffraction Night + Shub starts)

• Panel: Diffraction Night holographic ticket on localhost; venues in Slack; shareable themed bot templates; plushies debate. Eric/@trykarat: Karat banking/credit for creators → grow-the-business bank; AI useful when output is correct + verifiable; creator stack = brands / outreach / rates / content bots.
• ~443 BRB → Grok Bot for Founders (Shub, Growth Eng). Maturity curve to staff-of-bots; bots with own computers; founder jobs = focus, velocity, insight quality. Slide: “Do things that don't scale (but now you can actually scale them)” via routines/webhooks.

~450:00–465:00 (Shub — Close / Prod / Stalk)

• Founder workflows: close customers on autopilot; QA + product-change updates; adapt to competitors; ship feedback fast. Crew: Close / Prod / Stalk / Proto / YapBot.
• Close Bot: “prep the call” → HTML prep for Northwind Labs (people, goodwill UI bug opener, billing/usage, agenda). Compounds from call transcripts (Granola). Activation perk email example ($1000 secret perk on milestone — demo).
• Prod Bot: “daily rundown” for FlyLo — PRs + issues + live site walk with screenshots/video + shipped metrics.
• Starts Stalk Bot (sign up for competitors → feature rundown); clip cuts mid-explanation.

~465:00–480:00 (Shub wraps Founders)

• Stalk Bot: competitor pulse (Craft, Notion) with onboarding video teardowns; can email churned users for why they left. Proto Bot: own Grok Bot instance; pull feedback → ship in hours — example share-button “pop” request → cloud agent testing UI.
• Yap Bot: voice/style twin from email/Slack/iMessage; drafts for Close/Stalk. Misc Bot: catch-all (Drake album check via IG).
• Lessons: let bots run free / invest in course-correcting / think deep on delegation. Tips: browser use costly — prefer APIs/network; audit routines; trigger via webhooks not constant cron; voice bot, import cookies, group by expertise, auto-optimize routines; meta-bot to audit other bots’ routines.
• Close: “Steal Stalk Bot” QR. Thanks.

~480:00–495:00 (Shub Q&A)

• Contest slide flash again (Starbase trip) [unverified fine print]. Setup: inventory needs by domain; prefer expert bots over one mega CoS. Multi-device confusion improving. Enterprise determinism: have model write/run code/decision trees. Import helper bot “in a few days”; interim 1Password / Chrome cookies.
• Zoom/transcripts: don’t over-prescribe tools; Granola cited as easy on-device path. UI vs API speed: headless helps; API-level UI speed still hard. Token tips: group chat burns tokens (bots talk over each other) — prefer tagging; forget old context; pick models via Cursor/Grok Build.
• Marketplace: hand-audited rankings; probe bots with “what do you do / how.” Cross-account bot chat: not yet.

~495:00–510:00 (Shub wrap + Jenny)

• Shub wrap: prefer cloud over local (local “janky”); bots don’t share memory but share filesystem (one VM, per-bot instances); Grok Bot ↔ Cursor Cloud Agents first-class — Grok can QA/test/PRs; Cloud Agents for heavy model control. Contest reminder again. AV desync into panel.
• Guest Jenny (video, with baby): ~7 businesses + Jemmela Media; 22 Grok Bot employees [unverified]. Architecture: Master Chief CoS + Boxy (inbox) + Scribe (Whisper notes → delegate) + CFO/bookkeeping.
• Build ask: business in 72h / events pop-up. Jenny: SF venue hunt is brutal — build Scout bot (~100 pax warehouse/gallery, evening next month, catering rules: outside food / preferred vendors / kitchen).

~510:00–end (~525)

• Jenny: Chief-of-Staff + reverse-engineer an event producer (planner → staff bots). Personal: resale negotiation bots (Poshmark/Depop/Mercari). Prompt: SF senior event planner budget for 100–200 guests — venue, F&B, staffing, basic AV. “Permit / red tape” bot for SF regs; AI legal review ≈ second-year law student (as said). Build human process first, then automate. Guest list: Meta/IG top local followers. Screen share failed.
• Day 1 wrap (~523): Slack + Notion + GitHub + Vercel + V1 lander/dashboard stood up. Contest reminder. Goodbyes ~524:50 → end card.

Product pivots

Company / infra names as they appeared: Potato Lab; org shipbythursday; repos pop-up, thursday, company; domain shipbythurs.day; Slack #all-shipbythursday, #pr-reviews, #bug-reports, #marketing; bots Marky McMarkface, steve, grokpot, tater, Hashbrown, dr eggbot, Lead Capture, Operator Research, drop, Prioritizer; plugins pStack / pstack.

Guests / sessions

Host panel throughout: Matt Palmer, Lauren Tan, Roshan Sadanani (roles drifted on cam: CEO/CPO/CTO / Chief Potato Officer — treat as stream banter unless locked later).

Unverified / soft checklist

• Contest official post URL, eligibility, and prize fine print (Starship / Starbase trip; Hawthorne Rocket Factory runner-up; Sept 29 deadline) — verbal only on stream [unverified].
• Live viewer counts cited on air (~80k early; ~430k during Codie) [unverified].
• Replay player view count (~1.4M) [unverified].
• Bot-widget overlay peaks (~26 bots / 547 messages) [unverified].
• Codie audience/follower/investment/view claims; pStack PR volume (~2,500/mo); Lingxi “10 min” auto-fix / solo mobile build claims; PM “double-digit % of merged PRs”; Jenny “22 bots / 7 businesses” [unverified].
• Whether shipbythursday.com was acquired (chat sniping mentioned) [unverified].
• Final public product name & locked concept (multiple pivots; still soft at Day 1 close).
• Exact CEO title among hosts (fuzzy / joking across stretches).
• 1Password “secrets not exposed to SpaceX side” architecture claim [unverified].
• Demo funnel percentages (FlyLo search→fare ~33.9%) — demo data [unverified].

Key quotes (selected)

Only quotes clearly captured in scribe drops — not invented:

• Lauren: bots can get a Stripe “link card” / credit card in Grok Bot.
• Lauren (PM shift): typing feels like “casting spells.”
• Peter Yang: runs a one-person business on Grok Bot — “talk to my bots all day instead of humans.”
• Amrita (architecture): “One Grok Bot cannot access another Grok Bot’s computer.”
• Lander copy shown: “SPIN UP A STALL. OWN THE NIGHT.” / Night-market: “Spin up a stall. Own the night.”
• Codie parting (as captured): “if you build it, they will not come.”
• Kevin / PM session: “Grok Bot is an agent with a computer. It feels like texting a colleague.”
• PM demo: “These are colleagues with a computer.”
• Lingxi takeaways: treat bots like interns; automate one level up; start with success/fail feedback loops.
• Shub slide: “Do things that don't scale (but now you can actually scale them).”
• Matt: Oct 15 floated to ground comms; hero flow = group runs a pop-up (venue, staff, menu, plan, scheduling).
• Landing page scaffolded and deployed to Vercel in roughly a few minutes from Slack prompts (as shown on stream ~150–165).

Coverage checklist (clips vs notes)

Local clip archive exists for the Day 1 segments (not published). Notes status:

Compiled for public Gist update from Day 1 live note drops. Do not treat soft metrics or contest details as independently verified.