THE TL;DR ON TODAY’S EMAIL

• Humata.ai and how to build resilience to AI platform risks;

• A pragmatic approach to prioritizing features for AI MVPs;

• and more!

THIS WEEK’S HEADLINES

The best new resources for AI solopreneurs

No-code

• BrokenAtom launches a no-code app builder, visually great for data-intensive apps (link)

• Buildship brings AI-enabled no-code to backend builders (link)

AI & Data

• OpenAI announces the Assistant API and custom GPTs, further pushing the innovation edge towards customization of AI agents that specialize on specific tasks (link)

• Maruti launches beta for low-cost API for open-source LLMs (link)

Growth

• MassInbox offers cold email outreach as a service, for a fixed price (150k emails for 3995USD). Best suited for post-PMF solopreneurs with a solid funnel who are ready to scale their sales (link)

• Cotractable is a free legal document generator that takes one sentence prompts and writes contracts for you based on thousands of examples (link)

DEEP-DIVE: HUMATA.AI

How to build resilience to AI platform risks and build MVPs that scale

On the surface, Humata.ai is a chatbot that helps you ask questions to your PDF files.

Given recent features launched by OpenAI (i.e. the exact same feature that is core to Humata), many people think Humata’s end is near.

However, beneath the surface, there is a long-term vision that has given it a great shot at surviving the first few years in this noisy AI market.

In today’s case study, I’ll dive into:

• The appeal of the innovation frontier

• Building beyond one-dimensional version ones

• The takeaway for founders

The appeal of the innovation frontier

For the last 20 years, there has been a major problem for data-intensive enterprises: the inability to effectively manage and query unstructured data.

And when a daily problem for big enterprises has not been solved in more than two decades of technological development, it creates a big tension in the market.

In 2022, the release of LLMs like GPT, Bard and Llama changed the game, and this decade-old problem became finally solvable with a technology that works.

Cyrus Khajvandi, a serial entrepreneur with a series of successful AI and data ventures (including raising a 100M round) saw this opportunity and went for it.

However, how do you approach building ventures at the innovation frontier in the AI space, where most of the tech development is done by platform players, barrier to entry is lowered and the competitive field is extremely volatile?

I will tell you how Cyrus approached entering the market with his VC-backed venture, and what you can learn from it as a solo founder.

Building beyond one-dimensional version ones

In this first year of the LLM frenzy, thousands of AI startups have been launched and most of them are basic GPT wrappers for single specific use cases, or only offering a specific feature based on a core GPT offering - just think of the countless Tweet generators, generic chatbots and text summarizers out there.

It’s a risky bet as such one-dimensional offerings can easily become a commodity, or the platform players can integrate that offer in their core set.

However, the upside of jumping early on one of the market opportunities opened by LLMs in (sometimes) old problem spaces - like querying unstructured data - is undeniable.

Founders can very quickly capture market share and capital that is being deployed heavily on this new innovation frontier. What it takes is a solid V1 and good distribution strategy in the niche market.

However, building a venture that is resilient to market changes in the fast-moving AI space requires a sharp strategy to diversify the product or service offering from V1 to V2.

Humata.ai has a solid vision for what their future looks like, and they’re currently transitioning from what could be considered V1 to V2 of their product. Let’s look at what they are doing.

Their user-facing value proposition for V1 is one of the most basic use cases of LLMs - asking questions and receiving answers based on the contents of your PDF documents, with sources.

As of November 8th, this is still the first item in the FAQ section of Humata’s website:

After OpenAI’s recent introduction of Enterprise plans and the ability to upload PDF files directly in ChatGPT, this value proposition is clearly much less competitive.

One could say that Humata’s business model is in danger, right?

However, I would argue that Humata is actually well positioned to continue to thrive as they now move from their version one to their version two.

While their first selling point of enabling users to easily query PDF files is now less enticing and will become less relevant to market needs, it has served them well to capture early market traction (they have millions of users).

The focus on building a clear-cut, simple V1 is what allowed them to get early traction in a busy and noisy market.

Now, they can further invest in their vision for future versions, which was sitting behind the scenes. ARK’s long-term investment thesis on Humata (link) is not tied to their ability to query PDFs using GPT, but rather to the ability of doing so in a cost-effective way through a layer of metadata that allows for faster querying.

This was the vision of Cyrus, and is what will differentiate Humata’s future versions from the first offering, which is now also available on ChatGPT.

This might imply a future change in market positioning from hybrid B2B/B2C to mostly B2B and an increased focus on enterprise customers with complex unstructured data repositories. However, now Cyrus has a cool 3.5M in funding to do so, in addition to the revenue and customer base Humata already generated in the last year.

The closing statement in ARK’s investment thesis is clear: “Humata appears well-positioned to provide a leading document analytics platform that articulates with and enhances the use of the technology”.

In the future, the platform will extend the core features provided by OpenAI, and enhance them. Even though that’s not what brought them here.

The takeaway for founders

Building additional use-case specific features on top of the core service provided by AI platforms is critical to having a chance to product stickiness and sustained success. However, building these additional features might delay the launch to market significantly if founders want to have a fully fledged package at launch date.

For (solo) founders, it is typically not feasible to have a full set of both core features and differentiating features at launch. However, even established founders like Cyrus, are choosing to go to market with simple value propositions in their V1, and then later solidifying their competitive position with differentiation.

The AI market is just too volatile to justify spending a long time on building a polished V1.

When I talk to solo founders, three uncertainties often emerge:

• They think that their AI-enabled product is not ready for launch,

• They are worried that it’s too basic,

• They are afraid of the risk of the platform players kicking them out of the market.

However, speed of launch and speed of execution are the super power of founders in this market.

While OpenAI might still cannibalize your client base, they are slower at adapting to specific use cases than you are.

If you can build a solid early customer base and get decent MRR with V1, you can move your focus to V2 of the product as soon as possible. That’s where most of the differentiation should be built.

Get out to the market with a version one quickly. If it sticks, iterate towards V2, if not, get to the next idea.

My only recommendation is to spend an extra day while designing your V1 to think about what V2 could look like.

Set out a vision for a future, not based on the capabilities of LLMs today, but the potential capabilities of LLMs in 3 years from now.

It will help you think big picture from the get-go, and allow for space of mind even if OpenAI releases the next startup-killer update.

A PRAGMATIC APPROACH TO PRIORITIZING FEATURES FOR AI MVPS

We saw how Humata.ai went to market with a strong V1, and is now iterating towards their V2 - however, when you don’t have a V1 yet, how should you approach your product development?

If you’re in the ideation phase of your AI-enabled product and you don’t know which features to include in your version one, I bring you a pragmatic approach that you can test for yourself today.

Michael Seibel from YC states that you should:

• Keep your MVP simple,

• Time-box your specifications to weeks, not months,

• Write down your plans to prevent constant changes and create accountability for yourself,

• Cut unnecessary features after one week of development,

• Never obsess over making the MVP perfect.

My take: this is ever more relevant in the AI space, where the same platforms you leverage to build your products can take over your market even more easily than other competitors. You should not waste any time in building a polished product but rather you should go to market with the simplest version one you can build. Take Michael’s advice of revisiting your spec one week into the building process, and cut out all the features that you know will not be relevant based on the additional context you have after that first week.

I encourage you to watch Michael’s 10-minute speech here.

UP NEXT

In the next edition of this newsletter, I will share an example of success story from a solo founder that built a simple GPT-powered tool and left his 9-to-5 life behind, and is now focused on solidifying its place in the market.

In upcoming editions, I’ll share more resources that help you (solo) founders in find your niche, build your AI products and get them to market.

If you have specific questions, or would like me to go deeper on specific topics, let me know by replying to this email.