Y Combinator · Launch Video Breakdown: Hook, Pacing & Motion Design

Scene-by-scene timeline & spoken transcript

1. 0:00 – 0:03

   The Hook

   Founder Introduction & G-Stack Tease

   “Hey, I'm Garry Tan, President and CEO of Y Combinator. And I'm here to introduce G-Stack.”

   On screen

   (No on-screen text)

   Camera

   Medium shot of the founder, slightly zoomed in, maintaining eye contact with the viewer.

   Motion

   Subtle camera push-in, clean cuts between founder and product UI, with a quick transition to the product demo.

2. 0:03 – 0:19

   Product Reveal

   G-Stack: AI Engineering Team in Action

   “G-Stack is an open-source toolkit built by me that turns Claude Code into an AI engineering team. It's like having a team of specialists.”

   On screen

   garrytan/gbrowser-tailscale Open project Clone from URL Quick start garrylist analyze_txt gstack gstack-website mail-client code baku garrylist-agent easy-chromium-compiles kumo gbrain gbrowser openclaw-v1 tax-app Rio de Janeiro v4 /office-hours Opus 4.6 M Fast Thinking Plan Build 1099-INT... OK so the pain is real but the consequence is friction/annoyance, not actual penalties. The user has multiple bank accounts and every tax season has to manually track down all the 1099-INTs. They caught them all, but the process was painful. This is important. The pain is real but it's an annual annoyance, not a crisis. That changes the product calculus. Let me push harder on this before moving to Q2. Let me probe: how painful? How many hours? How many banks? And critically - is this a "I'll spend 2 hours once a year" pain or a "this ruins my entire February" pain? OK, so the pain is real but it's friction, not crisis. You didn't get hit with a penalty... you just burned hours tracking everything down. That's an honest answer and it matters. Let me push on this. Because "painful annual chore" is a different product than "thing that prevents IRS penalties." User input AWAITING RESPONSE How painful, specifically? APPROACH A: Gmail Scanner + Smart Checklist (Minimal Viable) Summary: Gmail OAuth → search for tax doc notification emails of which banks issued 1099-INTs with direct links to automation. User downloads manually. CPA matchmaking Effort: S (1-2 weekends) Risk: Low Pros: - Ships fast, validates demand immediately - No browser automation complexity - Gmail API is well-documented, OAuth is straightfor - Solves only half the problem (finding, not fetchin - User still has to log into each bank manually - Low defensibility — anyone can build a Gmail searc Cons: Reuses: Gmail API, standard OAuth2 flow Scroll to bottom Stack - Gmail + AI Browser Automation + CPA Marketplace RECOMMENDATION: Choose Approach C (CPA-First) because it solves the distribution problem. Completeness: A=4/10, B=8/10, C=7/10. Approach A is too thin — a Gmail search checklist is a feature, not a product. Approach B is the dream but it tries to solve everything at once (full automation + two-sided marketplace). Approach C is the wedge: CPAs bring you clients, and each CPA relationship is worth 50-200 taxpayers. The playbook: land 5 CPAs. Their clients get the Gmail scan + browser-assisted download flow. You charge the CPA $10-20/client. That's $500-4000 per CPA per tax season. You can always add the consumer-facing version later. Which approach? 1 Approach A: Gmail Scanner + Checklist (ship this weekend) 2 Approach B: Full Stack with browser automation (the dream) 3 Approach C: CPA-First (recommended — distribution built in) 4 Hybrid: start with A, add C's CPA layer next 0 B - but actually we B - but actually we could use the user browser interaction to skip Google OAuth entirely and just have the user open Gmail and a version of GStack Browser could just use Gmail to find the 1099's to search for automatically. Simultaneous to that, it could also ask the user what other banks they have. Also if they already have a CPA you can find out from the email, and if you're me, you probably already have a bunch of emails from your C You killed OAuth. The entire Google OAuth consent flow, the API keys, the scopes, the review process, the "this app wants to read your email" screen that scares people away, gone. The user is already logged into Gmail in their browser. The AI just reads it like a human assistant sitting next to them would. And you turned the CPA's nagging emails into the checklist. Your CPA already sent you an email saying "I need your 1099-INT from Chase and your W-2 from Acme Corp." That email IS the document request list. The AI reads the CPA's email, cross-references it against tax doc notification emails from banks, and says "Your CPA wants 5 things. I found 3 of them, let me go get them." The architecture is now: 1. GStack Browser opens Gmail (user already logged in, no OAuth) 2. AI searches inbox for tax doc notifications + CPA request emails 3. AI builds a unified checklist: "Your CPA needs X, Y, Z. Banks notified you about A, B, C. Overlap: these. Missing: these. What other banks do you have?" 4. AI navigates each bank portal (user logs in, AI takes over, downloads PDFs) 5. Package everything for the CPA. No backend. No API keys. Local-first. Browser automation only. Let me write this up as the design doc. 1. Prepare design doc met... eval "${-/claude/skills/gstack-siu... Write 102 lines garrytan-1099-int-collector-design-20250415-163838.n Error Cannot edit files outside allowed directories (/Users... Thinking These are the 3 directions I'll generate. Each takes ~60s, but I'll run them all in parallel so total time is ~60 seconds regardless. A) "Command Center" — Dense data table, status chips, all info at a glance B) "Friendly Progress" — Warm, progress-bar-driven, large cards, checkmarks C) "Split View" — CPA requests on left, found docs on right, visual matching RECOMMENDATION: Generate all 3. They're genuinely different approaches. Completeness: 8/10. 1 Generate all 3 — looks good 2 I want to change some concepts 3 Add more variants 0 Type something... Submit Bank of America 1099-INT CPA Found Download Wells Fargo 1099-INT Manual Found Open Citi 1099-INT Citi Missing Download Ally Bank 1099-INT Downsord Downloadt Download Discover 1099-INT Found Missing Download U.S. Bank 1099-INT Massaf Missing Download PNC 1099-INT Add Missing Add CPA REQUEST MATCH 3 of 5 items matched Pick What do you like/dislike? Option B Design direction B

   Camera

   Close-up on the product UI (code editor, terminal, and AI responses) with the founder's face in a circular inset. The UI pans and scrolls to reveal new information.

   Motion

   Dynamic UI scrolling and text reveal, simulating real-time interaction. The founder's inset provides a personal touch. Smooth transitions between different sections of the UI.

3. 0:19 – 0:24

   Feature Teaser

   G-Stack's Comprehensive Sprint Process

   “What is G-Stack? One install. 28 slash commands. Claude becomes your entire team. Each skill feeds the next. Nothing falls through the cracks.”

   On screen

   What is GStack? One install. 28 slash commands. Claude becomes your entire team. CEO who challenges your product scope Designer who builds design systems + variants Staff eng who catches production bugs before you QA who opens a real browser and tests your app Security officer who runs OWASP + STRIDE Release engineer who ships, monitors, and deploys 58K+ GitHub Stars 28 Slash Commands 17K LOC/Day (Garry) How I use it: plan mode fanatic I spend 80-90% of my time in /office-hours, /plan-ceo-review, and /autoplan. Everything about my app is figured out — especial plan mode. Then I let the agent build. Then /review catches what I missed. 35% of my code is tests. The model doesn't care abo badger don't care. Spending ~$4K/month on fast mode. Super worth it. t Process thing falls through the cracks. Design /design-* System, variants, HTML output Build Claude code Implementation with skills Review /review Staff-level bug catch Test /qa retro + /learn + /document-release → feeds back into next sprint The Sprint Process Each skill feeds the next. Nothing falls through the cracks. Think /office-hours Reframe the problem Plan /autoplan CEO + Eng review Design /design-* System, variants, HTML output Build /claude-code Implementation with skills Review /review Staff-level bug catch Test /qa Browser QA, regression tests Ship /ship PR, CI, deploy /retro + /learn + /document-release → feeds back into next sprint Why this matters for your startup: /office-hours asks the 6 forcing questions YC partners would ask. Challenges your assumptions before you write code. /plan-ceo-review finds the 10-star product hiding inside your request. 4 modes: expand, selective expand, hold, reduce. /plan-eng-review forces diagrams (sequence, state, data-flow) that surface hand-wavy planning before it costs you. /autoplan runs CEO + Design + Eng review automatically. Surfaces only taste decisions for your approval.

   Camera

   Animated infographic slides with text and icons, showcasing the G-Stack features and workflow. The founder's face remains in a circular inset.

   Motion

   Slide-in animations for text and graphic elements, highlighting key statistics and process steps. The founder's inset provides continuity.

4. 0:24 – 1309.1

   Call to Action

   Try G-Stack Now

   “G-Stack makes it easier to build software. Try it out.”

   On screen

   (No on-screen text)

   Camera

   Medium shot of the founder, looking directly at the camera, with a confident and encouraging expression.

   Motion

   Direct address to the audience, with a clear call to action. The founder's presence reinforces trust and authority.