Mentor Session XV: Selling Beats Building - Paul's Referral-Network Platform, Phone-Book Outbound, Give-First Communities, One Use Case Not a Hundred, and 'Know Thyself'

Harshit frames the moment: the building black box is open, so decide whether you wear the technical hat or the sales hat; executives wear the sales hat and hire engineers to scale what they prototype. He thought voice agents were commoditised, then signed one last week because the buyer had seen three vendors demoing recordings and his team had customised for the buyer's marketplace. The lesson is not the playbook but access: who do you already reach? Inbound (LinkedIn, social, brand) works in six to twelve months and should always run in the background. Outbound starts in your phone book and WhatsApp history: pass your idea through everyone you have chatted with, not as a pitch but as 'do you happen to know someone who would draw a lot of value from this?' - they answer as the customer or they connect you. Twenty messages a day, build the list, bake the habit, repeat next week; you have to become 'that sort of shameless person'. A generic capability (RAG, a chatbot, a voice agent) becomes saleable when applied to a vertical you can reach. Paid email outbound at $50-100 a month gives poor results by comparison.

Paul asks how to tell which products are saleable before investing too much love, then describes the one he thinks is: US networking groups of 10-30 people meet weekly in person to pass leads (the famous national one costs about $1,500 a member a year, with rules, and the morning is gone); his group meets online 8:30-9:30 and he has built the website that runs it - member contact information, referrals passed in a couple of clicks, follow-up and earned-business tracking, one member per industry, anchored by the real-estate agents and finance people who are the referral nodes. It has made his own group more effective and organised; he found no direct competitor; he plans a tenancy model so each group is private, and sees high-performing real-estate agents across Connecticut as the obvious buyers. Harshit's questions: what does it do, have you shown them, are there competitors? Then the strategy: be the power user of your own product and let people experience the value rather than be sold to - the way Praful's AIC Mastermind invites you to a session and you are signed up before you notice; the way Harshit gave away a resource playlist that was really a blog sign-up. Organise the meetups through it (he names Luma, luma.com); don't charge at first; get them 'stuck in the interface' chatting and scheduling. Research Skool, Luma and similar community platforms for ideas and to see whether the wheel is unique (he types a list into chat). Create one resource so valuable your own mind says 'I'd charge $500 for this' - and give it free as the offer; money comes after two or three months. Paul's plan in response: a video to entice people to a sample session, a trial with five to ten cohorts per group, then the per-group subscription. Praful adds the real-estate version of the same rule: 25 years of give-first.

Cody is building a giant MCP - 'OpenRouter for MCPs' - that other bots can plug into, and cannot get it tested or trusted outside hackathons; bots hit login walls and take website claims as fact. Harshit's diagnosis: too many features, no legible use case. That morning a senior Indian government official gave him two minutes with no introduction; he pitched the core in forty-five seconds. Firecrawl's MCP is easy to trust because it does one thing. For his own many-featured MCP (Glued) he picked one end use case - a performance marketer reallocating campaign budget between Instagram campaigns - recorded himself doing it, sent the Loom, and customers started asking 'can it also do...?' and selling the rest to themselves. So: list the five most important end use cases, define the customers even if it shrinks the TAM (there are no users yet), record them one by one, and test whether each is a real pain. Position the tool as a building block in the customer's bigger picture ('today we build a real-estate listing directory - the data comes through Firecrawl') rather than the picture itself. Keep a directory page for everything else. Cody's instinct to let 50-100 Outskill learners build verticals on it is 'a lot going on' - eliminate. On his AI-app-store question: there will be no app store; software is being quashed and money is made behind the scenes by APIs and MCPs that coding agents call (ElevenLabs, Higgsfield), so build vertical MCPs for specific audiences. On the 'check my balance, am I good for the holiday' agent: nobody connects their bank, travel and everything to your tool until you are OpenAI or Anthropic - a wrong assumption. On a JSON-schema capability registry: too low level; 'your head is in the codebase'. Prescription: one day, no code changes, brainstorm with AI in the same project on the single segment to pivot around, write the business plan. He sold a university vice-chancellor with a HyperFrames/Nano Banana video of a product that did not exist yet.

Praful recommends The One Thing (Gary Keller - the real-estate book with the domino stack and the two-rabbits opener, about 120 pages, one sitting) and will bring ten copies to Bangalore. Harshit admits the same pattern: sign a deal, then chase the next one instead of delivering; decide at New Year to focus on YouTube, get distracted by whatever turns up. Sarah's version, from a 1997 book on the new internet (Futurize Your Enterprise): don't say yes to everyone - people outside your target market fall into someone else's bucket; saying yes to all comers got her into ugly situations in the website business and the rule applies to AI work now. Harshit's close for the cohort: with so many playbooks learned, the phrase everything leads to is 'know thyself' - people who know themselves get answers from inside; people who don't have things happen to them. Pause, let others run in directions that are not your race, look sideways.

Seyaj is taking an umbrella company with several products to Web Summit Lisbon (Nov 9-13) with a demo day and VCs present, no pricing decided. Harshit does not attend many conferences but describes what he did for C4: build a skill/automation on the coding agent that reads the speakers' (and attendees', if listed) LinkedIn and X profiles - recent posts, the problems they discuss - and finds the intersection with your product, then make a small hacky prototype or video for exactly that. Give people something to take away from a two-minute conversation: a QR code that drops their own analysis onto their phone. From a Dubai crypto conference: the day sessions were half-empty, the after-party was where it happened; everyone shared contacts by Telegram QR, took a selfie with you and typed a note so they remembered who you were. Brainstorm forward: a QR tied to a dictation device that transcribes the conversation and sends a summary into the chat right there.

Vikram asks what base harness Harshit runs (Superpowers, Graphify and the like). Answer: he keeps deleting harnesses; one custom harness for deep development, otherwise simple automations, all on Codex on the Max plan. Spec-driven development kits (Superpowers, BMAD) are for software engineers with something already in production that must not break - planning, testing, review, refactor - and he uses them when adding a feature to a live product. For proof-of-concept work he dictates at length, corrects, and whenever he loves an output he says 'turn this into a skill: how we iterated, what I don't want (bloated scripts, em dashes)', and the next time he asks for a script the skill applies itself; 15-20 projects run this way. On the flood of terms - software factory, 'JEV' (an encoder/decision API the labs had internally), vibe coding, forward-deployed engineering (FDE: consultant + product manager + engineer, Palantir hires MS CS + MBA; not every engineer can hold a C-suite conversation) - stop obsessing: the terms are good for session titles, but the sessions end up teaching the same basics. 'There comes a time when you should stop listening and focus on one thing.'

Cody asks whether to continue SkillPort, a skills marketplace. Harshit: Vercel released skills.sh in a day because anything installed through npx registers there automatically and the real business is cloud services; directories and marketplaces bring traffic, but the next step is selling something, so first have the underlying business, then the marketplace - otherwise you will spend months chasing listings and make nothing, and people will compare you with skills.sh. Praful disagrees on framing (the platform itself is the value) and wants to fold it into a 'service piece' he is designing for cohorts (shared infrastructure to remove individual usage struggles) - to be discussed in the AIC Mastermind. Rajendra has 50-60 learning products and one he means to monetise, CauseOurClaims.com (OpenAI hackathon winner, going production-grade on Azure with domain testers). Harshit: focus on the one you are passionate about - V5 may fail, V20 may succeed; it is endurance; and stop talking tech - nobody cares about Azure vs AWS, only 'this customer had this problem, I solved it'. Use the habit of writing and sharing as automated SEO: daily posts on specific problems people search for in Google, Perplexity and ChatGPT so the product shows up; 'you're a very good lead-magnet guy - the leads need somewhere to go'.

The Bangalore immersion will be December 5 or 12 (Abhishek's message said 12; Harshit is trying for the 5th; decision within 48 hours); it runs from about 10 am to late, with founders invited, Harshit interviewing them, an after-party - nothing like the AIGF demo day at the Google office. Rajendra asks for a plan early (last time 25 people met the day before) and mentions a Jagriti Yatra idea. Town Hall is November 1, the last scheduled session. The 'Shekhar' (as heard) evals session was running in parallel today; recordings of it are not provided for now. After the course there are no office hours, but sprint and community events continue. An advanced Catalyst is not planned; Harshit may run a separate forward-deployed-engineering programme for software engineers. Sarah asks that Founder Stack guests talk about struggles, not just wins; Harshit will brief this weekend's guest. Weekend: Jarvis session Saturday, Founder Stack Sunday. Side notes: ElevenLabs hackathon group forming; Dubai AI Festival Oct 26-27.