As the commoditization of foundational AI models continues, the true battleground for innovation and value creation is shifting decisively to the application layer. 

At Behind Genius, we believe this transition isn't just about building applications—it's about enabling a new category of intelligent, autonomous agents that act with a level of independence previously unseen. 

This brings us to what we call Service as a Software, a concept where agentic AI acts as a service layer that not only executes tasks, but also makes decisions and optimizes workflows autonomously. Software evolves from static tool to active participant—one that understands context, learns from certain interactions, and ultimately delivers value by acting on behalf of the user.

A powerful example from our portfolio is Rooted Ag, which is pioneering the use of agentic AI in regenerative agriculture. Their voice agent acts as a digital agronomist, business consultant, and financial manager, providing farmers with tailored, data-driven insights in real time. By gathering data directly from farmers and integrating with weather data, soil testing results, historical performance, and more, Rooted Ag’s agent can autonomously adjust grazing plans, recommend optimal financing strategies, and even automate supply chain logistics. This autonomous, agentic approach not only boosts yield and reduces resource waste, but also empowers local farmers to focus on strategic decisions rather than getting bogged down in operational minutiae.

From manufacturing to agriculture, from enterprise to consumer products, the definition of software as a service will expand due to the continued acceleration of AI. 

The Role of Startups in the Continued Acceleration of AI 

In the three years since ChatGPT’s launch in 2022, AI has undergone an acceleration. In 2025, we now see the large foundation models becoming increasingly commoditized, driving down costs along a natural evolution of the technology’s cost curve. Recently, we saw the emergence of Deepseek R1, which will only increase the competition amongst the leading AI labs.

These advancements and the overall evolution of the industry have also redefined the entire AI value chain—pulling early stage value away from core, commoditized layers and toward the higher-value customization and application layers. 

With traditional AI infrastructure largely consolidated among public market giants, we feel that the true competitive advantage is emerging in how startups fine-tune, integrate, and deploy AI to address specific real-world challenges. 

The AI technology stack can be broken down into the following four key layers: infrastructure, model, customization, and application 

1. The Infrastructure Layer: As the foundational level, the infrastructure layer includes the hardware, cloud resources, and data ecosystems that power AI. In doing so, it ensures the heavy (but necessary) compute power and storage needed for training and deploying models.
   
   1. At this base layer, we think about chipmakers and hardware providers, such as Nvidia, AMD, TSMC, and Intel. Increasingly, despite export controls, China is also making progress in this space. Per a report by Goldman Sachs, Huawei is now making hardware comparable to Nvidia’s A100 chips which may lead them to capture international market share in the coming months and years.
   2. Beyond the hardware layer, we see Amazon Web Services (AWS), Google Cloud Platform (GCP), Microsoft Azure, and Oracle Cloud supply the massive cloud computing power and storage that are required by today’s models.
2. The Model Layer: At this level, we have the core AI models, namely frontier large language models. These models, due to their sheer size and volume, provide general-purpose capabilities that can then be applied across a range of tasks.
   
   1. We see this layer as the “brain” of modern AI systems. OpenAI, Google DeepMind, Anthropic, Meta, Mistral, and now Deepseek are key players at this layer.
   2. Despite the billions of dollars that have been deployed into training and developing models and particularly with the emergence of Deepseek, we feel that models are becoming increasingly commoditized, thereby significantly altering the structure of the value chain in AI (more on this later).
3. Customization/Fine-Tuning Layer: Here, the pre-trained models are able to be further refined and tailored to specific domains or applications. This involves adapting the model to particular datasets or use cases, enhancing performance in very niche areas.
   
   1. Open-source hubs like Hugging Face provide the tools and libraries to fine-tune and ultimately deploy models across a variety of diverse use cases.
   2. Azure’s OpenAI Service and Google Vertex AI are also cloud-based platforms that allow developers to easily customize large pre-trained models with their own data. Amazon’s Bedrock is also an emerging option at the customization layer.
4. Application Layer: The bottom layer leverages the outputs from the underlying layers to create end-user products and services; at the bottom of the stack is the application layer, where startups and innovators build differentiated solutions that address real-world problems, making this layer the primary source of value in AI investments.

‍

With respect to the model layer, as mentioned, we see increased levels of  competition, which positively leads to lower costs for consumers and developers, as well as elevated rates of adoption across both the consumer and enterprise customer profiles. 

As an example of this increased level of competition, per the Financial Times via Menlo Ventures, we see that Open AI has actually lost market share (dropping from controlling half the market to roughly a third), with Anthropic doubling its presence. While the traditional closed-sourced models are becoming diluted, we expect that open-source models will begin to proliferate.

As models commoditize, unless a catch-all winner emerges through a technical breakthrough, the market-share will continue to be distributed amongst a series of well-capitalized players.

Value Redistribution in the application layer.

As the basic layers of AI become standardized, higher-value opportunities lie in customizing AI for targeted applications and industries.

Why do we believe value will be found in the customization and application layers?

Let’s take an example in consumer tech. You can do a lot with ChatGPT, but many people don’t realize it. In contrast, specifically targeted applications allow the user to forgo broad AI education & address their specific needs - as a parent, caregiver, or in the workplace. 

Many of the earliest AI startups were easily outmaneuvered by incumbents who had distribution advantage, higher levels of resources, or an established brand name. However, with newer, more powerful models coupled with advances in hardware, the narrative has shifted. Innovations in open-source frameworks and enhanced customization tools also now allow developers to fine-tune models with unprecedented levels of precision. 

As a result, what were once seen as mere wrappers have evolved into critical components of the application layer, enabling companies to build tailored, efficient, and scalable AI solutions that directly address specific industry needs. This convergence of improved compute, advanced model capabilities, and robust customization options firmly places the application layer as the true source of value in today’s AI landscape.

Applied AI will have its moment due to technical advances.

Thanks to the rapid commoditization of foundational AI models, the prevalence of equally powerful open-source models, and significant breakthroughs in compute power and customization, applied AI is emerging as an exciting trend across an array of sectors, particularly in robotics and manufacturing. Why is this the case? 

• With models becoming more accessible and API call costs gradually declining, businesses can now leverage advanced multimodal AI systems to integrate and process diverse data types in real time. 
• This convergence enables a more holistic, human-like understanding of complex environments, which is particularly crucial for applications such as autonomous robotics and intelligent manufacturing systems. 
• These systems benefit from recent hardware advancements that reduce latency and improve processing speed, making it feasible to deploy AI in real-world, time-sensitive operations.
• In addition, the technical evolution in applied AI is enabling industries that have long relied on manual processes to modernize their operations. The powerful and potent integration of edge AI, advanced sensor networks, and real-time analytics is providing these sectors with the tools to automate a series of tasks, predict maintenance needs, and optimize workflows with unprecedented precision. 
  
  • For example, in manufacturing, AI-driven predictive maintenance systems can analyze data from sensors in real time to foresee equipment failures, while in robotics, advanced computer vision and sensor fusion allow for more adaptive and autonomous decision-making. 
  • These innovations not only streamline operations and reduce downtime but also create significant competitive advantages for companies ready to invest in the applied AI space.

Portfolio company Hum, formerly known as Coastal Carbon, exemplifies the potential of applied AI in the Earth observation data industry—a data-intensive vertical that is increasingly valuable for its ability to convert massive amounts of geospatial information into actionable insights. By developing a GPT-like system for Earth observation data, Hum is building an all-seeing, all-knowing agent capable of parsing and analyzing information across every square kilometer of the planet. 

Focused on critical sectors such as defense, energy, geospatial intelligence, government, and climate, Hum’s end-to-end solution addresses key challenges in data acquisition, preparation, and model application. With a robust sales pipeline and early traction—including a successful water security project in Texas and promising discussions with the National Geospatial Intelligence Agency—Hum’s rapid progress underscores the immense potential of applied AI to drive efficiency, reduce costs, and revolutionize decision-making in this high-growth market.

Active vs. Static Software: Defining Agentic AI

According to ChatGPT 4o, Agentic AI refers to artificial intelligence systems that exhibit a degree of autonomy, decision-making capability, and goal-directed behavior. Unlike narrow or task-specific AI, agentic AI operates more independently.

Agentic AI is rapidly emerging as one of the more exciting trends in the evolving AI landscape, promising to revolutionize how we interact with technology by automating complex tasks and streamlining daily workflows. As active, intelligent, autonomous agents become integrated into browser interfaces and enterprise systems, a northstar metric of success will be the time they save—transforming passive processes into seamless, efficient operations, largely free of human intervention. 

We predict heavy proliferation of agentic AI in both enterprise and consumer verticals in the coming months. 

• In the enterprise space, these agentic systems will likely work to streamline workflows by managing tasks such as scheduling meetings, processing large datasets for real-time analytics, and even handling customer service inquiries through intelligent virtual assistant
• On the consumer side, agentic AI will likely take the form of enhancing daily productivity, acting as personal assistants that can organize emails, book travel or dining accommodations, automate routine browsing tasks, and optimize smart home devices to create more efficient living environments. 

These use cases underscore the potential of agentic AI to deliver measurable time savings and improved operational efficiency across the board, bridging the gap between current capabilities and the promise of true AGI.

Rather than serving as a basic utility in operating systems, these next-generation agentic AIs are setting the stage for a future where autonomous, intelligent agents seamlessly integrate into our daily lives, dramatically enhancing productivity and redefining how we interact with technology.

As value shifts to the application layer, technical storytellers will remain a critical founder characteristic we look for. 

As a core part of our thesis, we believe:

1. Technical rigor is critical in sectors like manufacturing and robotics, where tangible operational improvements and cost savings serve as the backbone of measurable ROI.
2. Equally important is the skill of storytelling, which allows these technical founders to articulate complex ideas in an accessible manner and foster strong relationships with customers, investors, and team members. As AI applications increasingly intersect with everyday business operations—from automating routine tasks to driving innovation in traditional industries—the ability to present a vision that resonates on both a technical and emotional level becomes a powerful differentiator. 

By combining a solid engineering foundation with clear, persuasive communication, technical storytellers not only accelerate product adoption but also cultivate an ecosystem of trust and collaboration that is essential for long-term success in the dynamic world of AI. 

Conclusion

The evolution of AI from a narrowly focused tool in the form of a chatbot to intelligent, autonomous services is a massive shift that is redefining industries across the board. The transformation from traditional, static software applications to dynamic, agentic systems encapsulates the power of applied AI, turning software into an autonomous partner. As the foundational models become increasingly commoditized, the true differentiation lies in the ingenuity of the application-layer solutions, where startups can leverage rapid advancements in the infrastructure layer to solve complex challenges. 

Looking ahead, the promise of agentic AI is set to unlock unprecedented opportunities, empowering both traditional industries and emerging sectors to reimagine their operations. With technical breakthroughs, enhanced customization, and a renewed focus on storytelling by technical founders, the landscape is ripe for innovations that merge human intuition with machine precision. The journey from mere automation to intelligent orchestration is poised to redefine business models and establish a new benchmark for efficiency and impact. As we stand on the brink of this transformative era, where the synthesis of technology, strategy, and narrative will drive the next wave of economic growth and societal progress.

‍