Discussing Superduper with Max Tee

Rod Rivera and Maxson Tee discussed SuperDuper, a young startup that aims to simplify the process of building and deploying AI applications. The company provides a layer on top of existing databases that enables developers to easily integrate machine learning and AI functionalities. They compared SuperDuper to competitors like MindsDB and LangChain, discussing its potential market, monetization strategies, and challenges in the current business environment.

Takeaways

1. SuperDuper's unique selling point is its simplicity and use of vanilla Python, making it accessible to Python developers.
2. The startup faces competition from more established players like MindsDB, which has raised significant funding and has a larger user base.
3. Potential monetization strategies include a cloud-based deployment solution with usage-based pricing, similar to Vercel's model.
4. SuperDuper could target both small teams/freelancers and enterprise clients, but may need to develop additional features for enterprise adoption.
5. The company's partnership with MongoDB Ventures could provide a significant advantage in terms of distribution and credibility.
6. Building and nurturing a strong developer community is crucial for SuperDuper's growth and adoption.
7. In the current funding environment focused on profitability, SuperDuper may face challenges but could attract investors if they can prove rapid value and market traction.
8. To differentiate itself, SuperDuper should focus on ease of use and potentially develop enterprise-specific features like enhanced explainability and governance tools.
9. The key to success for SuperDuper lies in lowering the barriers to entry for AI application development and deployment.
10. The startup's ultimate goal should be to become the "Ruby on Rails of AI applications," establishing itself as the go-to framework for AI development.

Episode Transcript

Introduction to SuperDuper

Rod Rivera: Welcome to our second episode of AI Pro Ducks. As always, I'm joined by Max, a veteran in the VC and innovation space. How are you today, Max?

Maxson Tee: I'm doing well, thank you. Looking forward to the weekend. How about you?

Rod Rivera: Indeed, the weekend is just around the corner. But before we wrap up the week, we have an exciting discussion about something that sounds like a database but isn't one - SuperDuper. Have you heard of them, Max?

Maxson Tee: Yes, I have. I must say, I love the name. It's interesting that they're not really a database, but rather trying to make your existing database "super duper," as we've heard from the founders. I find their approach fascinating - they're doing something really cool by making it easier for people to deploy AI applications, especially for those who don't necessarily have all the AI knowledge. They're enabling users to build AI applications on top of traditional databases. What are your thoughts, Rod?

Understanding SuperDuper

Rod Rivera: Exactly. For those unfamiliar with SuperDuper, it's a very young startup that went live and started promoting itself late last year. The idea behind it is that they add a layer on top of your database to enable machine learning and AI applications.

The challenge they're addressing is that building any type of AI application, whether it's classical machine learning or generative AI, is difficult. It typically requires many roles and libraries to glue all the code together to create something usable. This process is time-consuming and complicated. SuperDuper offers this "glue" and simplicity to enable a single developer to do what would typically require a full team.

Max, you've looked into their business model and team. What's your impression? How does it align with your framework for validating startups?

Team Analysis and Market Potential

Maxson Tee: It's interesting. When you talk about all the different elements needed to glue an AI application together, I think of super glue - one thing that pulls everything together and allows you to deploy applications much faster.

Regarding the team, I think they have excellent experience in this space. Timo, for instance, built a search application for e-commerce that was doing very well. He then created an AI-based innovation lab studio before going full-time on SuperDuper.

Similarly, the CTO, Duncan, has an interesting background. He didn't start coding until he was 23, which shows it's never too late to begin. From there, he fell in love with AI and machine learning applications and just kept building.

I feel both of them have domain knowledge in this space, both from an application perspective and a developer perspective. One key insight they seem to have gained from their past experiences is that building an AI application is painful. Their goal appears to be alleviating that pain to achieve functional AI applications.

Another point I agree with is their belief that 99% of AI applications don't need to scale like crazy or be extremely performant - they just need to be good enough to improve the solution.

Rod, as a builder and AI founder yourself, what are your thoughts on this? Have you seen alternatives to what SuperDuper is offering?

Comparison with Alternatives

Rod Rivera: Yes, the closest alternative I see to SuperDuper is a company called MindsDB. They've been around a bit longer and were part of Y Combinator. They're quite popular in terms of users and investor interest. Their GitHub repository has over 19,000 stars, and they've raised more than $50 million in funding.

The products look similar, but there are some notable differences. With MindsDB, you need to learn a sort of meta-language to use it. They work with SQL, the language for database querying, but it's not exactly one-to-one. It's similar but has some differences.

In contrast, SuperDuper uses 100% vanilla Python. If you already know Python, in theory, you just need to learn SuperDuper's functions, and that's it. MindsDB's approach has some advantages, though. It enables non-technical or semi-technical profiles to work with it because many analysts in marketing or finance departments know some SQL. They can get started and use the AI functionality by writing SQL queries.

With SuperDuper, you need to know Python, which is a bit more limiting as only Python developers can use it easily. However, for existing Python developers, SuperDuper's adoption is super simple.

Market Size and Monetization

Maxson Tee: That's very interesting. Statista reports that in 2022, there were over 10 million Python developers, making it the second most used programming language. The ability to enable non-coders to deploy solutions is intriguing, but I believe people can learn Python if they can learn SQL.

I think SuperDuper chose Python because it's the language most commonly used for data science and ML-related tasks. They seem to be targeting this group, making it easier and faster for them to deploy applications.

My next question is about market size. This could potentially be a very big market, especially if it becomes the lingua franca for AI application development. But how do they monetize? They're open-source, which is great for adoption, but what's their business model? Based on your conversation with the founders and your background in this space, what do you think their monetization strategy might be?

Rod Rivera: The software remains open-source, and it's still very early. I imagine their current focus is on adoption and growing the community, so monetization might not be at the top of their priority list. However, if you go to their website, they already have a waiting list for a cloud product.

I can imagine a model similar to Vercel, which is famous in the front-end web development space. Vercel offers a framework called Next.js, which has become the standard for building user interfaces on the web. It's free, but Vercel monetizes by allowing users to deploy their websites to the Vercel cloud, handling serving, caching, and efficiency.

I can envision something similar for SuperDuper, where you develop your application and then deploy it on their cloud. They could charge based on the number of requests, with pricing scaled to usage. If your model is popular and widely used within a company, you'd pay more; if it's more of an internal tool with less usage, you'd pay less. It would likely be a pay-as-you-go, usage-based pricing model for model deployment.

Target Customers and Enterprise Adoption

Maxson Tee: That's interesting. In terms of usage-based pricing, do you think they'll target smaller startups or aim for larger corporates as well? Or could it be both?

Rod Rivera: I think the goal of any company in the tech space is always to target the enterprise. It's unlikely they'd say no to enterprise contracts, big names, and logos. However, looking at the product itself, I do think it's a very nice fit for small teams, solo developers, and startup teams. There are many freelancers out there who are asked to develop AI applications for various use cases. For these one-person shops that need to handle everything, SuperDuper is the perfect solution. It takes care of a lot of the heavy lifting, allowing them to focus on developing logic and machine learning models while using SuperDuper to connect to data and deploy the application.

Maxson Tee: That's super interesting. I'm thinking about SMEs (Small and Medium Enterprises). There are many SMEs in the world that have deployed some sort of databases to keep track of their data. If you're able to run a consultancy and build SuperDuper on top, using it to deploy AI applications for these old SME applications, that could be powerful. SMEs don't necessarily need the same level of performance as large corporations like Visa or Mastercard that handle millions of transactions.

It's really enabling smaller players to develop applications without the usual headaches. It's allowing AI developers to focus on AI while everything else is taken care of for them. It's a bit like joining a football academy where you go in to play football, and everything else - food, lodging, etc. - is taken care of for you.

From a corporate standpoint, particularly in banking which I understand best, the ability to explain an AI model is becoming increasingly important due to regulatory scrutiny. You need to be able to show why a model is producing a particular output. There's a whole aspect of model governance to consider. Making deployment easier also needs to make the models more explainable. This could be an interesting angle if they want to pursue the enterprise route.

Given that they're open-source today, explainability is mostly relevant for enterprise customers. This could be another monetization feature they offer to enterprises - making it easier to deploy and then making it easier to explain the models deployed on top of the database. What are your thoughts on that?

Enterprise Features and Differentiation

Rod Rivera: Yes, I see that perhaps at the moment, all the necessary features for enterprise adoption might not be there. As you mentioned, access control, governance, and similar features are essential for enterprise customers and might not be fully developed yet. But I think that as the framework evolves, and also to differentiate itself not only from MindsDB but also from frameworks like LangChain (which has become almost the default way to build AI apps), they'll need to offer these additional features.

You can argue that SuperDuper also lets you do classic machine learning use cases and is broader than something like LangChain. But at the same time, LangChain is a name that everyone in AI knows about, that companies understand and are asking employees or potential employees to develop skills around.

As a result, it might be that LangChain ends up going in a similar direction as SuperDuper. One way for SuperDuper to differentiate itself is to really start targeting these enterprise use cases. They could say, "Hey, to monetize our product with this SuperDuper cloud, we'll offer everything from SOC 2, HIPAA, and other certifications, all the way to having all these access control and governance features that regulated industries such as finance would require."

Maxson Tee: Yeah, that's interesting. Even within large corporates, the AI governance processes are not necessarily there yet. It's either they over-govern or under-govern. So I think there's definitely an opportunity to break into this space.

You touched on something really interesting about how other players in the space like MindsDB and LangChain might be able to go after similar use cases as SuperDuper. That brings to mind the question of how defensible SuperDuper is. As you said, if they become the lingua franca, then it becomes a user lock-in. But other than that, we'd love to hear your thoughts as well. How defensible is SuperDuper, and what do they need to do from this perspective?

Defensibility and Future Strategy

Rod Rivera: That's something that can be debated quite a lot. The more it matures and the more users it gains, yes, there is a lock-in effect. Another thing that makes it defensible to a certain extent is the level of integration. If you go to their website, you'll see many integrations with various database vendors and other technologies. This takes quite some time to develop.

You might be able to create a proof of concept similar to SuperDuper over a weekend, but getting all these integrations in place, maintaining them, and ensuring they're robust is a whole different story. It's quite complicated to get right and takes a significant amount of time. So as SuperDuper matures, this gap with any new entrants will increase.

But I'm also thinking about how they can differentiate themselves from MindsDB, which has been in the market longer and has raised a significantly larger amount of funds. In these cases, Max, what would you advise? SuperDuper is much newer, smaller, and more nimble. How can it compete against these more entrenched players in the market?

Maxson Tee: Interesting. I think my general sense, based on my observation of what they're doing, is that they've obviously got some funding from MongoDB Ventures. That in itself is super interesting because MongoDB is one of the most famous databases out there. If they can become built into MongoDB, so that every time someone uses MongoDB, SuperDuper is available, that would be a significant advantage. Getting a strategic partner as a distribution channel would be super interesting because it kind of supercharges your ability to reach clients without having to knock on everyone's doors. It also brings some credibility, which allows you to roll out your solution much quicker.

I think separately, it depends on which areas they are trying to tackle as they go. The current funding environment will probably allow them to build quicker, but they will have to be very disciplined about where they decide to build features. They are probably going to speak to a lot of their community members to get input. So building up the community as they have done so far would be super interesting.

So I guess those are the two main things: some sort of strategic partnership from a distribution perspective, and then building with the community. I'm thinking about how to build and how to sell via strategic partnerships to reach some of the bigger players and bring in larger deals, as you mentioned. And on the other side, continue to work with the community to build up all these key functionalities that people are looking for, so that the deployment and usage of it will become more widespread. If they truly want to be the "Ruby on Rails of AI applications," then the community build is super important.

Rod Rivera: From a VC perspective, we're in 2024, and since 2023, we've been talking about this being the year of profitability and revenue. How do you judge companies like SuperDuper that are focused first on this community-building approach, where monetization is not yet fully there or non-existent, and where the product itself is provided for free as an open-source product? How do you judge them? Do you think it's a good moment for them? Or might they face some challenges in getting interest from investors? How can they thrive in this environment?

Maxson Tee: I think ultimately the proof point is that they are indirectly targeting the same type of users as LangChain and MindsDB. Whoever can win the hearts and minds of those developers is where I feel the investors will focus.

Developing the community of developers and making it easier for people to deploy AI solutions would be key. It's almost coming back to what Y Combinator says: build something people want. I think that's one thing to think about. And yes, the funding environment is probably going to be very challenging. I think it comes down to how they can utilize the solution to prove value as fast as possible.

The story we've been telling so far and its ability to eventually become like a Vercel or Ruby on Rails for AI - I think that would get investors excited. Personally, I feel like it's a venture bet, and the question is how good the team is and whether they can bring it to a venture-like return.

Because we're in the year of profitability, it's probably good to think about the monetization strategy. We obviously conjectured that they might go down the Vercel path, but they might be able to come up with something more interesting that allows them to provide value to others while extracting enough value to become a big company.

Even as an investor, we probably care a lot about the traction, the team, and fundamentally, is it a vitamin or is it a painkiller? Are we solving a big problem here? Which in my head, it feels like they are.

Rod Rivera: If you were to have one piece of advice for the SuperDuper team or anyone in this space building a similar type of framework, what would that be, Max?

Concluding Advice

Maxson Tee: Make it easy for people to deploy AI applications. If I can use it to deploy an AI application to replace my own job, I think that's the holy grail. If you make it easy for me to deploy AI applications myself, then let's go. Like we said last week, a lot of existing software solutions will eventually deploy AI. The market is big enough, and building something that makes it easier for people to deploy AI is crucial. I think SuperDuper is probably on the right path already. They have the right mission.

Rod Rivera: I see it similarly. Making it easy might really be their moat. Of course, adding new integrations makes a lot of sense, but new integrations also mean complexity. If we can lower the threshold for people to adopt AI, which is where we are at the moment, and SuperDuper positions itself as the easiest way to create these AI applications, then they can really capture this community. From there, they have a moat that's impossible to overcome.

Later, if they decide to do an enterprise product or a cloud solution, they'll have this very faithful, very enthusiastic, and growing community of new developers creating AI applications. That's everything they need. So I do hope that they focus on this usability and ease of use, enabling the building of AI applications within minutes.

Maxson Tee: Yeah, lowering barriers to entry is what comes to mind. Democratizing the ability to build AI applications, to use the buzzword of the last century. So, yeah. Cool. Thank you very much. This is exciting.

Rod Rivera: Excellent, Max. In our next segment, we'll be talking with SuperDuper's two founders, Timo and Duncan. Great talking to you, Max. Let's reconnect next week.