Back to Basics: What to Address According to Where in the AI Stack You're Building

Application Layer and AI Infrastructure Edition

‍

Sentiment towards AI is complex. As I’ve had conversations with researchers, builders, company executives, and other investors in the space, sentiments have ranged broadly from highly optimistic to pragmatic to quite skeptical. Yet, there's always a consensus: AI will significantly alter their landscape. The ambiguity, however, lies in the details—where and how this transformation will unfold and what existing solutions will be displaced.

At the same time, it seems that every startup is now an AI startup and every corporate leader has been asked about their “AI strategy”. Many of these executives are grappling with identifying opportunities, formulating effective strategies, and trying to understand what it even means to implement those. As a result, I believe enterprise adoption of AI will be slower than what we’re collectively expecting. This may cause some gap in conviction from investors as reality catches up with expectations a year or two from now. Startups that go out to raise during that time frame will likely face this disenchantment. What’s going to matter in that time is the value a startup is providing to customers.

I believe for founders, the outlook remains positive despite the high expectations set for AI startups. This current wave represents a significant shift in the technological foundation and the opportunity over the long term is immense. As mentioned above, success, however, demands more than just integrating AI; it requires a clear value proposition for customers and investors.

For early-stage founders, here’s what I believe is most important to address today that will set you up for long term success and better investor conversations based on where in the AI stack you’re building:

Application Layer

Why does a startup capture the value here?

How to address:

• Understand why your startup will win. “We’ll build it faster because we’re a startup,” isn’t a robust enough strategy. Large companies (who have access to more data and resources) are also aiming to maximize AI's value.
• I’m most impressed by founders that can outline the playbook of incumbents and competitors in their respective industries. Detail your opportunity against both incumbents and new rivals, focusing on workflow, user experience, and feature superiority.
• Wedge into a non-obvious piece of the market. If you can identify an overlooked market segment or customer base that existing solutions aren’t solving for, it can allow you to scale up and get data while they’re not paying attention.
• Know where the user experience differentiation matters. Perplexity AI went horizontal despite nearly everyone telling CEO Aravind Srinivas to go vertical (except Marc Andreessen, as he noted in a recent interview). Aravind understood where he could deliver value to the user and what needed to be captured in that user experience to do so.  

Is this defensible?

How to address:

• Have a clear data strategy. If there's no strong technical moat for what you've built, you’ll need to demonstrate that you have access to proprietary data or explain how you’re going survive long enough to gather proprietary data that would in time lead to more robust defensibility.
• Formulate distribution advantage. With the pace at which companies are tackling similar problem spaces, you need to show are you going to acquire customers that someone else isn’t going to be able to easily replicate.
• Recruit well. Good talent and team can provide differentiation. Don’t underestimate the importance of building your team out to fill skill gaps. An investor should look across the team and think “Wow, no one else can tackle this problem except this group.”

Are you addressing a genuine need?

How to address:

• Define the use case. Horizontal applications struggle with this the most when the product is generalizable and able to do many different things. So much so, that even customers are unsure what use case to apply it to. Doesn’t mean you need to be vertical but defining a killer use case is crucial for repeatable go-to-market and sales strategies. If the problem you're solving changes at every organization, so will your ICP within the organization which will slow things down.
• Prove this through traction.

AI Infrastructure

Won’t AWS, Azure, GCP, etc. just build this?

How to address:

• Steer away from building a temporarily missing feature to a platform. On the infra side, I see this most with startups trying to build features that the computing platforms or model providers are lacking today. This can be limiting as you ultimately lack control of the direction of the underlying platform but it’s likely the larger platform will catch up to filling the gap in their product. There's little stopping the platform itself from developing a similar feature, often with more resources and direct access to the user base. This puts your startup in direct competition with the platform, which is a tough battle to win.
• Understand where these platforms are not advantaged to providing a superior user experience and articulate why it matters.

Is this a feature or a platform?

How to address:

• If you’re dependent on partnerships or other platforms to sell to a customer, you may need to rethink how to capture more of the problem. Monitoring, ops, and observability tooling seem to struggle with this more than other parts of the stack.
• Your product should be venture-scale on its own.

If you’re open-source, how are you monetizing?

How to address:

• Develop a well-aligned pricing strategy and understand the right way to price what you’re building. Whether it’s paid support, paid premium content, paid access, hosting, etc. ensure it’s priced in a way that makes sense for your product and metrics you’re wanting to drive.
• Do the work to start understanding this early on.  
• Don’t over index on GitHub stars alone to carry you through. More Series A investors are wanting to see traction through revenue. Start early on establishing  a sales motion and get your product to paying customers.

Are you building in a category that will be around five years from now?

How to address:

• Assume that the things that are 70-80% solved today will likely be fully solved within the next five years, with architectures to support them.
• This is less about having a perfect crystal ball answer and precise vision on how everything will unfold. This is more around showing you understand how things are evolving and are building directionally in the right way to take advantage of innovations in model structures and systems as they will undoubtably change.

Understanding these nuances and addressing them head-on is not just beneficial; it's essential for long-term success of your startup.

Have good answers to these questions? Let's talk! hg@flyingfish.vc