Open Source Conversational AI - Berkeley 2024

[Music] I'm going to share a story with you all today and uh what I am hoping for you guys to take away from this story is two things first off like how the startup Journey can be very you know undetermined and and you know find its own way in a sense second of all how to navigate open source more specifically how to navigate the commercialization of Open Source so that's going to be the the theme of of the story today this all started with me myself and my co-founder Ismael back home in Sweden the the city of gothenberg so it's a high-tech City very much like you know SF and and the B area we have self-driving cars on the streets we have you know EV cars we have you know battery manufacturing Etc and uh but the startup scene is not very hot so ismile and myself we were running for the last you know six or so years a Consulting business software Consulting life was good life was comfortable right we were nowhere near the startup scene of you know Silicon Valley that was just you know fiction for us and one day we got kind of bored so we started building some cool AI products and we got to know a lot of people in the AI community and what we observed was that a lot of people in this community was Reinventing the wheel and we said okay we have some tech right that we have built for our own products it was was essentially Tech to run AI agents in a very easy way so about 10 months ago this was last summer we op sourced the tech we had and we had no commercial plan whatsoever we were not doing a startup we were just sharing what we had built and uh so that's why we called it super agent right doesn't sound like a company just sounds like a cool thing you put out there we open sourced it on GitHub and we just invited people yeah you know check it out let's let's just describe what super agent is super agent is a platform for running any kind of workflow using AI agents everything is abstracted away behind an API so you don't need to know anything about AI you know llms rag or any type of these AI Concepts you just need to know how to use a rest API so we open source this thing but this thing goes Goes Bananas gets a lot of traction and people start using it for all kinds of stuff we have big Banks using it we have a bank with like 990,000 employees who has deployed super agent internally for all kinds of weird stuff and we were like okay what's going on you know something is something is happening right and we start to have contributors from all over the world so we have like people working fulltime writing code for super agent which is crazy we don't pay them but but you know there's an interest built around this thing people are getting a lot of value from it and therefore they are incentivized to contribute we were on to something we couldn't continue our existing path um with our consulting company so we just stopped everything we were doing and just worked on super agent and we said we need to go to Silicon Valley to figure this thing out right take a flight we come over here again we have no idea what we are doing we meet with all these investors and we realize okay there's like a market for investing in open source companies uh we get accepted into y combinator we turn down the other investors we go and do y combinator and we get into a room with a bunch of other open-source projects just like us who are trying to build startups the main learning is there is no one way of commercializing an open source project right turns out there are many ways of doing this so it all depends on who your customer is what type of business model you are going for how difficult is it to run your project yourself on your infrastructure what type of value are you creating what part of the software stack are you addressing there are so many different dimensions and what we did and I hope this is valuable for you all our first commercialization attempt was to try to sell a cloud offering to developers because that made sense because developers are the ones who uses our GitHub project makes sense right turns out that this was very difficult because the developers wanted to run it on their own cloud or on their own you know infrastructure right so we had like a project Market fit but we had nowhere near product Market fit so what we decided to do next is that we identified a second customer group among our users that was very different from the developers and this was business users who were building a very specific type of AI agents and that was AI agents for web research so they were like spinning up AI agents to do research on you know customers markets like very businessyoutube [Music]