I’m Anirudh Kamath.
I recently wrote about my decision to forego my acceptance into Y Combinator’s S24 batch.
I consider myself a product-minded engineer, and I’m trying to make something people want. I’ve been a software engineer at FAANG, an early engineer at series B through series F unicorn startups, and most recently a founding engineer and first hire at a seed-stage company.
My industry experience spans consumer, e-commerce, B2B SaaS, and FAANG with various eng roles primarily in AI/ML, MLOps, and full-stack.
Most software applications boil down to creating, retrieving, updating, and deleting (CRUD) data. Consider Twitter as a simple example: it allows you to create a post, retrieve tweets in a feed, update your tweets, and delete your tweets.
Taking the time to build a frontend allows you to build rich and fully-customized user experiences like infinite scrolls, lazy loading, etc. However, it comes at substantial financial and time costs to incentivize users to use your app. Even if you have a significant, game-changing app, you’re going to spend a lot of time in developing the frontend, incentivizing users to download your app, and building/deploying push notifications to retain them.
However, building a startup is simply the art of removing confounding variables in an experiment to zone in on what people want and how you can deliver. Discord, Slack, Telegram, and FB Messenger all have rich chatbot interfaces that allow you to mimic most interactions that you would see in a . Making use of these rich interfaces to provide compelling user experiences will yield to a very compelling use case.
I don’t mean chatbots in the traditional sense of natural language to natural language, rather just mimicking basic frontend operations with the underrated richness of a bot UI.
Midjourney generates $200m in revenue as a Discord bot. BonkBot generates $10m a month as a Telegram bot. Partiful grew by being a website and seamlessly incorporating into your group chat with a simple link — it has an app now, but is still a fully functioning product independent of its app.
There were still problems to solve before ChatGPT came around. A lot of those problems still exist, and a good chunk can be solved without AI.
AI will proliferate every industry, much like the internet has. If you can build a unicorn business independently of AI, you can 10x it at-will with AI. AI is also still so nascent and changing so fast, that placing any core bets directly on the future of AI are just highly unlikely to make a dent. If you find any patterns to abstract, chances are thousands of other people have also found that pattern, and the market will be extremely saturated in a matter of weeks.
Just 2-3 years ago, chatbots were unreliable at scale because they were routinely forming unintelligible nonsense. The key language abstractions were missing, and we’re quick to forget Microsoft’s AI chatbot that started spewing Nazi propaganda.
We quickly went from being wildly impressed that a chatbot could even form full sentences to now having a chatbot on a Chevy dealership in Watsonville, CA that can run simulations for complex partial differential equations.
In 2024, AI enables semi-technical people to create software like never before. Instead of having to ask StackOverflow about how to configure Kafka for your specific PubSub architecture, you can instead just tell ChatGPT you want to setup real-time data and follow step-by-step instructions.