Spotlights
From the Founder Catalyst: Service-as-Software, powered by AI Agents
Read more
This interview is part of the Decibel OSS Spotlight series where we showcase founders of fast-growing community-led projects that are solving really unique problems and experiencing strong community adoption.
Sudip Chakrabarti spoke to Asim Shrestha and Adam Watkins, two of the three co-creators of AgentGPT, the open source project that allows one to build, configure and deploy Autonomous AI agents. Name your own custom AI and have it embark on any goal imaginable, and AgentGPT will attempt to reach the goal by thinking of tasks to do, executing them, and learning from the results.
Asim and Adam shared with us their inspiration behind creating AgentGPT and their vision to make it a widely adopted project.
Asim: We are all from Vancouver, and before this project, I studied at Simon Fraser University. I met Adam during our software engineering class, where we worked together to build a custom Sudoku game. That experience sparked our interest in hackathons, and we participated in a number of those, achieving a fair amount of success. Later, we both joined the BC Cancer Research Center, working on machine learning and building pipelines to bring ML results into the clinical environment. We also collaborated on an open source project for translating segmentation results for medical images into the DICOM format. Afterwards, I joined AWS to work on distributed systems and RDS, while Adam went to become a tech lead at Article, a fast-growing startup in direct-to-consumer furniture retail. We both enjoy working on really challenging technical problems and have always bonded doing that.
Adam: AgentGPT was born more out of a frustration than a motivation - a frustration that Asim and I shared from our prior jobs. Having been software engineers, we both have been on call numerous times and have had to wake up at ungodly hours to resolve critical issues, the fix for which would often be fairly simple like rebooting a machine or cleaning up the /tmp folder. About 7 months ago, when we started tinkering with AI, we realized the power of LLMs in building deep visibility into complex software and tracing customer issues all the way to actual software bugs. So, we started to build a “co-pilot for SREs” that would help automate troubleshooting and remediation for software issues. However, around this time we saw projects like Auto-GPT and BabyAGI take off and realized that there was a much broader need for automation. So, we did a micro-pivot and started building a more generic tool for automation, something that was far more easy to use, far more accessible and open source. That is how AgentGPT was born.
Asim: Our grand vision is to automate both personal and business processes. Currently, our platform is used by both consumers and businesses for automating tasks like market research, analysis, and travel itinerary planning. Our goal is to provide people with more time for activities they truly cherish, such as spending time with family and going on vacations. By leveraging AI to boost productivity, individuals can enjoy more free time. We want AgentGPT to be the tool that any user, irrespective of their technical skills, can use to automate tasks in their personal and professional lives that can be automated.
Adam: I actually have a personal story that captures very well how we see AgentGPT making a huge difference in someone’s life. When we launched AgentGPT, I happened to be backpacking in Europe at that time. The project took off like a rocket - we had over 100,000 users in the first week itself - and I was caught up in scaling the platform, waking up at 3 am every day making sure we did not go broke because of the massive infrastructure costs we were incurring. I did not have any time to plan my own itinerary. So, I became the first big user of AgentGPT, and it saved my trip. I'd ask it, "Hey, I'm in Berlin for just one day. What can I see between 5 pm and 10 pm?" And guess what? It would give me a list of museums open that day, tell me how to book them, recommend awesome restaurants to try afterward, and even suggest hotels and ways to get around. If I had relied on Google searches, it would have taken me hours to figure all that out. But with AgentGPT, I could simply ask one question and plan my entire day in Berlin. It was a lifesaver, really. Thanks to that, I could focus on scaling the platform successfully while still managing to see a few things during my trip!
Adam: The breakthrough here came from being first to market and being easily usable more so than just being technically complex. The two things that stand out the most for me are:
Asim: The adoption of AgentGPT, in one word, has been simply phenomenal. Take GitHub stars for example - the last open source project I had created had about a 100 GitHub stars, that too after two years of work. I think Adam had another project that crossed 1000 stars. With AgentGPT, however, we have already crossed 24,000 stars and are closing in on 50 contributors, all just in 2.5 months! The other metrics we track are: Discord members (23,000+) and DAUs (10,000+). We also closely track our OpenAI bill, which is ~$1,000 a day now, just to make sure we do not go bankrupt. Finally, we recently launched a paid subscription plan and we of course closely track the number of paid subscribers.
Adam: Speaking of paid subscribers, I remember the night we launched paid subscription we were super pumped when the first customer signed up. We all then went to bed and woke up next morning to 70-80 emails from annoyed customers who could not sign up because we had a bug in the signup flow. It was stressful, but at the same time overwhelming to see people wanting to pay for what we were building!
Asim: We find that people pay for our paid subscription for a couple of different reasons. First, some of our users want to use AgentGPT but don’t want to download the open source project and self-host it. So, they pay for our hosted version of AgentGPT. Another reason people pay for is because they want integrations to other tools they use, e.g., Slack, Notion, Google Docs, Google Sheets, etc. Finally, there are a handful of people who pay simply because they have gotten value from the open source project and want to support it.
Asim: Whether we should open source AgentGPT or not was never a question for us. We all have been open source creators in the past and have experienced first hand the power of a thriving community. For example, one of our open source contributors took the time to internationalize AgentGPT; so, AgentGPT is now available in 20 other languages, something that has increased the reach of the project tremendously and was only possible because of the community.
Adam: I also believe that open source is one of the best ways, if not the best, to test whether a project provides real value to its users. Without that, there is no business and hence, no company to be built.
Asim: We definitely weren't prepared for the rapid success that came our way, and it led to some mistakes along the journey. But we're actively working on fixing and improving things now. Looking back, here are three things I wish we had done better. First is having a clear project roadmap that is publicly shared with the community. That way, everyone knows exactly what we're focusing on as contributors and what areas others can lend a hand with. We have now taken the time to document our plans and goals, both in our project documentation and on GitHub. Second, we have put significant effort into making the code more modular and plugin based, so that contributors can add or change a module in isolation from the rest of the code without having to learn a wide range of technology stacks to make a single change. It is like offering building blocks for the community to play with and create something amazing. Finally, we've recognized the importance of thorough documentation. We've come a long way from the early days when running the app was a cumbersome process, requiring users to figure out complex configurations and keys. Now, we have a setup script that streamlines everything, allowing contributors to quickly jump in and get started. We understand the value of making it easier and more enjoyable for our contributors to join our project, and we're continuously improving to ensure a smooth and collaborative experience.
Adam: I’d add two other thoughts here. First, to build a successful open source community, it's crucial to actively engage with your community. When people join our Discord, they're not just seeking help; they want to exchange ideas and discuss AI, agents and automation. As founders, we need to recognize this and foster a vibrant and engaging community. Second, a mistake we had made was not being proactive enough in onboarding new contributors. As our project grew rapidly, we received a massive number of pull requests. Our focus shifted towards tackling significant challenges, leaving less time for onboarding and mentoring new contributors. However, I've come to realize that they are the cornerstones of any successful open source project. In hindsight, I wish we had prioritized onboarding and spent more time with these contributors. It's important to make it easy for new contributors to join and actively engage with them. They bring fresh perspectives and enthusiasm, which are essential for the growth and success of the project.
Asim: I have two projects that I absolutely admire. First one is llama.cpp. It has paved the way for open source language models and their accessibility on various devices, even at the edge., which is a really big deal. The other one is the Dolphin emulator which has always held special significance for me. It allowed me to experience games that I couldn't afford as a child. The dedicated community behind Dolphin has nurtured its growth and given several of the really old games a second shot.
Adam: I have always liked open source projects that solve deep technical problems. Hence, I really like Supabase, which has had an incredible run since its inception at YC. The other one I have a lot of respect for is the FastAPI framework. We use it for AgentGPT and it has cut our time to first byte from 10 seconds down to less than a second, solving a lot of usability problems for us.
Adam: My biggest piece of advice is create something that solves a problem for yourself. If what you create solves a problem for you, it likely solves the same problem for others too. And if it solves it well, then it will surely be very successful. So don't be afraid to share your project if it is solving a problem you have, because you never know how many people might be facing the same problem every single day.
Asim: I completely agree with what Adam just said and I would add that it is really important to be consistent. Once you have an idea of what problem you want to solve, just keep working at it - a little bit every day goes a long way toward building a hugely successful project.
