Fund Raising Playbooks for AI Startups: Mining YC's Request for Startups, Pressure-Testing an Idea with Gstack, Building a 12-Slide Pitch Deck with an Agent, and What Investors and Co-Founders Actually Look For

Harshad opens with Y Combinator because of the volume of public material it publishes. YC runs several batches a year, he says fewer than one percent of applicants get in, and the standard deal he quotes is 500,000 USD for a stake in the form of a SAFE (figures are his, covered again at 1:46). Its Request for Startups page (ycombinator.com/rfs) is refreshed about quarterly and each entry is written by a different partner: a personalised AI tutor (the 'Primer' from Neal Stephenson's The Diamond Age), the future of American defense, a cloud for small software, multiplayer AI, AI for low-pesticide agriculture, AI-native discovery engines and AI-native service companies. If you are building something close to an entry, you can apply under that heading; if you have no idea yet, the page is a free brainstorm from partners who watch the market. The entry he picks for the demo is AI-native service companies, new because YC historically avoided service businesses as unscalable. The phrase is service-as-software: the company does the work (insurance, banking, accounting, tax and audit, compliance, healthcare administration) instead of selling a tool, because total spend on services dwarfs spend on software and much of it is already outsourced. He links it to Sarah Tavel's (Benchmark general partner) 2023 essay 'Sell work, not software'. His positioning point: software is a commodity, accomplished work is what customers pay for.

Gary Tan (YC's CEO, as spoken) published two open-source repositories: Gbrain, described as his opinionated Hermes/OpenClaw-style agent brain (his notes, markdown files, to-dos, updated almost daily) and Gstack, the setup he uses to act as CEO, designer, engineering manager and release manager. Harshad recommends Gstack for fundraising because it contains the office-hours material YC partners use: a startup diagnostic template and a flow that goes to and fro with a founder the way a YC partner would. Installation is one instruction to the agent: paste the repo link and say 'install Gstack' (here for Codex on his Mac), and tell it to wait for the idea. When he later gives it the idea, the agent announces it is using the Gstack office-hours workflow, asks about the customer workflow, challenges weak points and recommends what to show. He is explicit about limits: customers are the only real validation ('if they are paying, the product is validated'); Gstack helps clarify and challenge but should not be relied on fully. He adds that you can build your own stack the same way and that YC's application now asks for skills as part of the profile (he later finds that field removed). Separately, he shows a private YC knowledge base he holds from his own batch (founder talks, investor updates, fundraising index), says he will share it with the cohort and asks them to keep it internal; nothing from it is in this record.

Asked how to present an AI-native service idea, the agent recommends twelve slides for an 8-10 minute walk-through with evidence in an appendix: what we do, the customer's problem by role, the service delivered, one real job end to end, evidence and progress, why now, pricing and delivery economics, initial market and expansion, how we win the first customers (go-to-market), why customers choose us over alternatives, the team and the ask. Harshad frames it with real decks: Airbnb's original (problem, price, solution, market validation, market size) and Uber's on SlideShare, then settles on Sequoia's template (available on Figma) over YC's, which he says suits younger founders and the YC application rather than institutional investors. The agent produced an editable PowerPoint plus a matching PDF in a cream, wine and black style; his rules: avoid dark mode (it looks cool but most decks are dark, so light differentiates), keep to twelve slides, put the main points on slide two because investors may give you five minutes, and show the product or a customer, since a demo or a paying user is a form of validation. He then repeats the exercise for his real company, Kenzai Intelligence (a decision layer / AI coworker platform that absorbs a company's knowledge base and surfaces decisions; the deployed customer turns WhatsApp threads into deployment and invoice decisions through MCPs and the Google Workspace CLI; target buyers with 20M+ ARR). He corrects the agent on camera: it mentioned a competitor that raised 32M USD (remove it), used its own design language (point it at the company's storybook and brand guidelines), used text instead of infographics, lacked a TAM slide, and printed the 20M ARR target on a slide (say it aloud instead). His closing TAM slide reads 10,000 qualifying companies at a 100,000 USD contract, a 1B USD opportunity (his example numbers).

Harshad draws the money chain on screen. Limited partners (he says 'liquidity providers' - family offices, founders with exits, countries) put capital into a VC firm; the firm is answerable to them and must show year-on-year returns. Inside the firm sit partners, general partners and a managing partner; a newly hired partner has to prove an eye for early companies, and the whole job is to deploy capital. His practical advice: get a referral to or pitch the newest partners, because they have something to prove, and build the deck to answer the question 'is the market large enough that a small share is still a big outcome?' That is the TAM (total addressable market): one percent of a small market is hard and small, one percent of a large one is large, so investors reward a credible, well-calculated TAM and judge how you calculated it. He admits that in his experience poorly designed decks sometimes raise the most, so clarity and a bullseye TAM matter more than design, but the product, team and co-founders still have to be good. He also describes how YC partners rotate and that politics exists, and presents it as human nature, from four years of observation.

Asked how to find a co-founder, Harshad points to YC's own co-founder matching platform and to applying even without a finished product (he says YC takes companies at any stage and has seen series A companies apply). Investors look for someone technical and someone who sells, with complementary backgrounds who 'click', and want the origin story of the pair; he says three is the sweet spot, two is common (the Collison brothers) and solo founders are now possible with agents. The application includes a founder video; he plays the famous DoorDash video (four co-founders, over 100 interviews, a first month with 150 paying customers and over 10,000 USD of sales, started as delivery drivers themselves) and mentions Airbnb's. Other fields he scrolls: company in under 50 characters, demo link and password, what the company will make, where you live and will be based, how you decide valuation, progress and users, competitors, how you will make money, legal entity. On valuation he explains the standard YC pre-seed instrument is a SAFE (simple agreement for future equity) on a post-money valuation cap, so there is no valuation at the start; his rule is to take a SAFE of around 250,000 USD and see how the next round clarifies the number. The next batch he gives is winter 2027 with an application deadline of 2 November. He remembered an 'add your skills' field in the application; on screen he finds it gone. On splits: equal, he says, because unequal splits cause conflict that shows up once series A or B money arrives and investors dislike it; titles matter less than equity ('no ego').

In the Q&A and stories Harshad repeats one theme: leverage comes from customers, revenue and product. Funding pays for runway (the 'winter'), hires and marketing; it is a chicken-and-egg problem, because investors back products that already have customers, so crack the product first. Practise: YC founders rehearse a pitch a hundred times before demo day, where the room includes top investors. An anecdote he tells to show that confidence matters: a young founder who acted as though he did not need the money said he was happy with 2M USD, and investors queued up. He also believes smaller raises can produce better products than very large ones ('comfortable with cash in the bank'). IP protection: a VC once took a full code-base walk-through and later invested in a competing AI-interview company, so his rule is to walk in with leverage and sign an NDA first; a participant adds that sending yourself a sealed letter timestamps an idea. Choosing investors: it is like a marriage, they sit on your board and have decision rights, so screen them. He is also critical of YC's recent direction (funding gambling-type apps) and says not to rely on YC alone. Against raising: 100 percent of a 20M USD company beats 10 percent of a 100M one, indie hackers run several products, and he cites Base44 (acquired by Wix for over 50M USD) and OpenClaw (Peter Steinberger, acquired by OpenAI) as solo-founder outcomes. For US investors he suggests a Delaware C corp, optionally via Stripe Atlas (about 500 USD). He closes by warning he is 'not a finished product' and that he too is still learning.

Before the main content Shivani reminds the cohort that two weeks remain: the next week holds the town hall (fill the form in the WhatsApp group to present your product, final touches allowed in the meantime) and the last session follows. The in-person immersion is confirmed for 5 December in or near Whitefield, Bangalore, probably at a Hilton; a one-day networking event with panel discussions and no project presentations, not a two-day event. Details are to be finalised in the first or second week of November and posted. Harshad adds he cannot paste links in the Zoom chat (a participant relays them) and will share the YC and fundraising folders afterwards.