The Rise of the App Layer
30 seconds summary
Since the launch of GPT3 we have been witnessing the dawn of a new era in technology - one that bears striking similarities to the early days of the Internet.
In the rapidly evolving landscape of generative artificial intelligence (generative AI), we're seeing new companies emerging in the so called app layer of the technological stack. This phenomenon echos the journey of the World Wide Web from the early days of web pages to the rise of SaaS apps.
In particular we're seeing a number of startups evolving from a simple skin on top of a sophisticated Large Language Model (LLM), to solving user problems through proprietary technology and expertise, building a sustainable competitive advantage.
Using the development of the Internet as a reference, we argue that AI is going to be a transformative force leading the emergence of a new generation of great app-layer companies.
At Crosscourt we're excited about leading the charge in investing in and coaching startups that are leading the charge in this transformation.
The First Wave: Generic Applications and the Protocol Layer
Just as the early Internet saw a proliferation of basic websites with limited utility, the first wave of AI applications has been characterized by generic chatbots and simple text-to-image generators. These tools, while impressive in their novelty, have often fallen short in terms of accuracy, reliability, and real-world utility.
Whilst it is common for new technological waves to go through a "trough of disillusionment" after an initial phase of excitement, it's crucial to recognize that this first wave was not about building user-facing applications - it was mostly about establishing the foundational protocol layer for AI.
Much like TCP/IP for the Internet, Large Language Models (LLMs), trained on vast amounts of Internet data, are the protocol layer of the AI era. They provide a foundational layer upon which successful apps can be built.
This is where the second wave comes in, focusing on use-case specialization.
The Second Wave: Specialized Apps
As we move beyond the protocol layer, we're entering the second wave of AI development. This phase is characterized by increased specialization.
Specialization includes all the complex systems, proprietary workflows and domain-specific intelligence that make an AI operate proficiently and solve specific user needs. In the context of AI, this requires sophisticated engineering, including custom architectures, specialized training pipelines, domain-specific prompting strategies, and fine-tuned models that can reliably deliver high-value outcomes for specific industries.
Specialized applications require purpose-built user interfaces that address their specific use cases. While the success of ChatGPT has led some to believe that text is the universal UI for AI applications, this oversimplifies the reality. Text-based interfaces excel in certain scenarios, particularly where written communication is central to the task. However, many applications, such as shopping assistants, require rich visual interfaces with product images, comparison tools, and interactive elements to support effective decision-making. In these cases, AI should enhance rather than replace traditional UI patterns, potentially by generating or augmenting visual content while maintaining familiar user experiences. One way this could be done would be to acquire a high-fidelity 3D image of the user and have the AI fit the clothing on the user's digital twin.
"The success of ChatGPT has led some to believe that text is the universal UI for AI applications. This oversimplifies the reality"
This evolution towards specialized apps that solve user problems mirrors the development of the internet ecosystem.
Consider Google: they didn't succeed merely by building one of the first internet search engines - they developed intricate ranking algorithms, crawler systems, and data processing pipelines that made Search uniquely powerful. Similarly, Amazon built its initial competitive advantage not just by selling books on the web, but by developing complex logistics systems, recommendation engines, powerful UIs, and inventory management capabilities that created a superior customer experience.
The main objections being levied against Generative AI apps
Critics have raised several key concerns about investing in the generative AI application layer.
First, there's a perception that these applications lack defensible intellectual property, as many are simple chatbots built through basic prompting and function calling. With frameworks like LlamaIndex, LangChain, or CrewAI, one could theoretically build such applications in days or even hours.
Second, there's fear that LLM providers themselves will dominate the application layer, especially as they venture into reasoning and General Artificial Intelligence, as evidenced by GPT-4 o1.
Third, many believe incumbent players hold an insurmountable advantage due to their vast repositories of user data, which they can leverage to develop AI-driven products.
Why app-layer generative AI companies will be successful
At Crosscourt, we see things differently:
A sustainable competitive advantage is being built at the application layer. Creating reliable, accurate, and safe AI systems for specific domains requires sophisticated engineering, carefully crafted system prompts, and extensive training. Companies can build sustainable competitive advantages through this specialized expertise and by creating a data flywheel that further improves later versions of their products.
LLM providers won't dominate every domain. While they may excel in general-purpose applications (e.g. search), specific industries require specialized user experiences and deep domain knowledge. As applications mature, opportunities exist to build competitive advantages through traditional means, e.g. process stickiness in enterprise SaaS or network effects in multi-party systems.
Incumbents don't necessarily have an advantage in new or previously undigitized verticals. Additionally, AI companies can operate at a meta-layer, orchestrating existing services and enabling users to focus on strategic decisions rather than tactical execution. This opens up a new layer in the value chain where new startups can thrive. Finally, many established companies lack the agility to adapt to AI's fundamentally different user experience paradigm, where AI acts as an agent rather than a tool.
Crosscourt is embracing this new wave
We're seeing these trends play out among our portfolio companies. For instance:
Traverse3D: A company we've partnered with since pre-seed has been building AI that can understand the physical world and can therefore be used to train humans in industries such as defense, construction, and mining.
Wand AI: Another company we backed early on has been developing specialized AI applications to enhance enterprise analytics. A side-by-side comparison of Wand's capabilities relative to ChatGPT shows the significant amount of expertise and process power Wand has built in this specific use case, making their product superior.
Rad AI: Creating AI solutions tailored for radiology centers to alleviate the workload of busy radiologists, demonstrating how domain-specific knowledge can be integrated with AI to create powerful, targeted tools.
These companies are building layers of complexity into their models to improve reliability, safety, and specificity for their particular use cases. They are also building powerful user interfaces that fit the specific needs of their use cases. This work is what allows these powerful tools to solve real-world problems effectively and ethically.
Long-term Competitive Advantage: Product-User Symbiosis
As the AI ecosystem matures, a further and very powerful long-term advantage lies in creating products that adapt to and grow with the user.
This advantage stems from AI systems that develop a nuanced understanding of the user's life experiences, previous interactions, current context, and emotional state. By incorporating these elements, AI can offer a more personalized, trustworthy, and enriching experience.
Consider, for instance, an AI assistant in healthcare. Over time, it could learn the specific terminology, protocols, and preferences of individual healthcare providers or institutions. This accumulated knowledge transforms the AI from a generic tool into an indispensable, personalized asset. The result is a symbiotic relationship between the user and the AI, where each interaction strengthens their connection and increases the system's value.
In this emerging AI ecosystem, the companies that successfully develop these personalized, learning, and context-aware systems are likely to establish the strongest competitive positions. They're not merely offering a static product; instead, they're providing an evolving, intelligent service that becomes more valuable and integrated into the user's life or workflow, with each interaction.
"Companies that successfully develop personalized, learning, and context-aware systems are likely to establish the strongest competitive positions"
The key to success in this domain will be striking the right balance between personalization and privacy, ensuring that these AI systems enhance user experience without compromising ethical standards and using the power of data to fundamentally benefit users.
Conclusion: The AI Opportunity
The parallels between the AI revolution and the Internet revolution are clear. We're moving from a phase of generic, often underwhelming applications to a new era of specialized, highly capable and personalized AI-powered tools and services, built on a solid protocol layer.
At Crosscourt, we're excited to be at the forefront of this transformation. We believe that the true potential of AI lies not in replacing human intelligence, but in augmenting it - creating tools that enhance our capabilities, streamline our workflows, and unlock new realms of creativity and problem-solving.
As we continue to invest in and incubate companies in this space, we're looking for those that understand the importance of building app-layer companies - teams that can build great products that solve a need users deeply care about.
The AI revolution is just beginning, and like the Internet before it, it will reshape every industry and aspect of our lives. If you are an entrepreneur looking to build new products using AI, the question is not whether AI will transform the World, but how you'll harness its power to create unique value for users.