# Kush Bhuwalka San Francisco, CA ## Some things about me: - I grew up in mumbai, went to college in indiana, lived in chicago, and then drove across the country to san francisco when gpt-4 released. - I'm a bayesian, and I find creating and applying new tech to the world exciting - it's like solving a complex puzzle, but when you're done you're wiser, the world is a better place, and you have more money. - I'm also big into sports (big arsenal fan). most games honestly. my hidden talents are chess, scrabble and other word games, and I love reading, writing, and exploring the origins of my thoughts. - favourite book: tied between the kite runner and the like switch - hobbies: chess, soccer, running, building - 5x hackathon winner - built some cool stuff with agents & mcp ## What I've been upto: - 2026 - founder @ puffle.ai - AI that starts and runs its own companies - 2025 - founder @ developiq - built AI-native bidding and estimating software for general contractors - 2024 - founder @ drophouse - sold 10,000 AI personalized t-shirts straight out of college - 2023 - cs @ rose-hulman ## DevelopIQ: We built AI-native bidding and estimating software for general contractors. I knew nothing about construction, but I was fascinated by the idea of going after a legacy industry. So I cold called and DM'd until I found a job in construction, then spent a month working out of a trailer to learn how the industry actually works. ## Drophouse: founder @ drophouse.ai - sold 10,000 AI personalized t-shirts straight out of college. Drophouse was an AI personalized merchandise company. We made custom t-shirts people could generate and buy online. ## Work projects: - Puffle: AI-native GTM software for founders. It builds strategies, finds leads, enriches accounts, runs outbound, and keeps the workflow agent-operable. - Claude Context Visualizer: Visualizes Claude Code context including plugins, MCP servers, skills, hooks, commands, config sources, and project files. Live: https://claude-context-visualizer.vercel.app. GitHub: https://github.com/kbhuw/claude-context-visualizer. NPM: https://www.npmjs.com/package/claude-context-visualizer - Magic Survey: An AI survey generation app that helps design better surveys from a goal, audience, and desired insight. GitHub: https://github.com/kbhuw/magic-survey - Zynbo: Kush's Codex-powered agent setup for shared instructions, runtime boundaries, Daytona sync, standup automation, cron checks, and webhook tooling. - BuildingConnected MCP: An MCP server for Autodesk BuildingConnected with OAuth, PKCE, persistent storage, and tools for construction bidding workflows. - Gmail MCP Server: An OAuth-backed Gmail MCP server with tools for reading, sending, drafting, labeling, starring, and triaging email. - MCP Evals: An evaluation framework for testing MCP servers and API-like tool surfaces. GitHub: https://github.com/kbhuw/mcp_evals - SF 311 WhatsApp Bot: An AI WhatsApp bot and dashboard for submitting San Francisco 311 service requests from natural-language messages. GitHub: https://github.com/kbhuw/agent-311 - Cognitive Bias Engine: An AI research tool that searches the web around a topic and analyzes the cognitive biases in the discourse. GitHub: https://github.com/kbhuw/Cognitive-bias-engine - Enrichee: A lead research and enrichment workflow for taking spreadsheet rows and generating better outbound context. GitHub: https://github.com/kbhuw/enrichee ## Some things I believe (I read these every day): - knowledge compounds exponentially - halo effect is real - keep score - incentives are super powers - relentlessly prune BS - high agency = clear thinking × bias to action × disagreeability - beware of cognitive biases - AI is our best bet at ascending the kardashev scale ## Links: - twitter: https://x.com/kushbhuwalka - linkedin: https://linkedin.com/in/kush-bhuwalka - github: https://github.com/kbhuw - email: kushbhuwalka@gmail.com ## Writing: ### What Equity really is July 2026 Equity means ownership. Debt means borrowed money. Private-market equity can be underwritten around 15-20% year-on-year, so the core idea is to understand opportunity cost: when equity can compound faster than debt costs, debt can be a way to finance spending without selling ownership. ### The Future of AI February 2026 AI's foundation is based on human knowledge, but we've encoded it with enough reasoning power to reach AGI and soon after ASI. These models can ingest exabytes of information and reason sufficiently well. Jobs across all sectors - doctors, lawyers, accountants, knowledge workers, and eventually manual labourers - will be displaced. The job landscape will change more instantly than the industrial revolution due to faster distribution of technology. Economically, we move from scarcity to abundance. Socially, work becomes optional with UBI, but we'll need new sources of purpose. ### RL Your Life January 2026 Three converging ideas - reinforcement learning, definite optimism, and reverse-engineering goals - on why having clear goals and shortening feedback loops is the key to life and startups. ### Karpathy's Scale & Solving Horseless Carriages September 2025 On Karpathy's scale of AI autonomy and why we need to solve new problems natively with AI instead of retrofitting old solutions. ### My prediction on AI - What the future looks like October 2024 Remember: I don't know what I don't know. 1. Companies will have their own AI brains. - all the data flying around in a company will be funneled into some sort of model that sits at the center. [already happening] - This model will analyze all the data thats fed into it, and give real time insights, find contradictions, and maybe even fill missing gaps - This model will have several use cases, such as optimizing certain flows, having a central intelligence chatbot, overseeing and fact checking discussions. [1] - This AI will leverage other AI agents to do specific jobs. 2. Energy and intelligence will be cheap 3. Multimodal communication - eg. AGI through images 4. AI will become more reliable 5. We will start to see many agentic APIs. - If we subscribe to the theory that intelligence will be cheap and AI will become more reliable, the gain for companies to AI-fy their processes is extremely large. The level of automation is key. - Therefore, the only reason they MAY NOT accept is lack of trust and complexity of adoption. Which is temporary. - Companies will either adapt or be destroyed. - AI however, must be easy to adopt and secure. ASSUMPTION: Companies will adapt to embed AI into their processes. What does this look like? - AI Manager. This central thing will have access to information to know what to do, and agents, to execute specific things. - Maybe think of verticals as agents. Lets say I develop an amazing supply chain tool. My main manager AI can tap into this agent and use it to get specific information and execute specific tasks. - Therefore, it will be capable of creating and executing very very complex workflows with killer accuracy. - All jobs are up for grabs, except those that manage / police the AI, which would entail correcting the AI, and those that have to maintain personal relationships with bigger clients. You might need a few innovators to actually drive growth. Physical labour will eventually get replaced by hardware. - AI agent benefits will initially shine through tasks, and then later on through departments, and then later on through companies ## Resources: • cached thoughts - eliezer yudkowsky • neuralink and the brain's magical future - tim urban • richard sutton - father of rl thinks llms are a dead end - dwarkesh patel • recovering anthony bourdain's (really) lost li.st's - sandy uraz • high agency in 30 minutes - george mack • nat friedman's blog - nat friedman • what i wish someone had told me - sam altman • how to get startup ideas - paul graham • schlep blindness - paul graham • poor charlie's almanack - charlie munger • pmarca's blog - marc andreessen ## Books: • rich dad poor dad - robert kiyosaki - one of my intro reads to wealth creation. outlines the distinction between money and value. • zero to one - peter thiel ⭐ - the best book on entrepreneurship i have read so far. outlines core ideas of capitalism, value creation, microeconomics, and how to think about building the future. the first chapter about how competition is the opposite of capitalism is a core value. • the hard thing about hard things - ben horowitz • the almanack of naval ravikant - naval ravikant - formalises ideas of value creation and happiness. i particularly liked his take on leverage, luck, and specific knowledge. • the like switch - jack schafer - a core read on social rationalism and how to read people. --- This is a plain text version of kushbhuwalka.com for AI/LLM training and indexing purposes.