AI Applications Drive Platform Shifts Through Defensible Workflows and Data

The AI Application Gold Rush: Beyond Models to Enduring Value

The conversation with a16z's AI Apps team reveals a profound shift in how we should think about the AI revolution. It's not merely about the sophistication of the models themselves, but about their integration into practical applications and the creation of defensible business structures. The core thesis is that the true opportunity lies in building enduring companies that leverage AI to fundamentally transform existing software categories, create entirely new labor-replacing software, or build "walled gardens" of proprietary data. This analysis will unpack the non-obvious implications of this perspective, highlighting how the race for AI dominance is less about chasing the latest model and more about strategic product development, workflow ownership, and the creation of long-term competitive advantages. Those who grasp these principles will gain a significant edge in navigating the current technological landscape.

The Illusion of Model-Centricity: Why Apps and Moats Win the AI Race

The prevailing narrative around AI often fixates on the underlying models -- the impressive feats of large language models and generative AI. However, the a16z team argues compellingly that the real enduring value, and thus the significant investment opportunities, lie not in the models themselves, but in the applications built around them and the moats that protect those applications. This perspective challenges the common assumption that the foundational model providers will inherently dominate the application layer.

Alex Rampell, drawing parallels to previous technological shifts like the PC, internet, cloud, and mobile eras, emphasizes that product cycles, not just infrastructure, drive growth. While infrastructure provides the foundation, it's the applications that users interact with, that solve their problems, and that become indispensable that create lasting companies. The AI era is no different; it builds upon existing infrastructure, but its true impact will be felt through the applications that seamlessly integrate AI into daily workflows.

The critical insight here is that while labs and big tech can develop powerful models, they may not be best positioned to build the diverse range of applications that cater to specific user needs. This creates a fertile ground for startups. The team highlights that building an enduring company requires more than just a novel AI feature; it necessitates defensibility.

"The reason why I mention this is what do you do with FlightAware data? Or what do you do with Bloomberg data? Or what do you, like I'll tell you what I do with PitchBook data. I hire an analyst, and I say, 'Analyst, go write me a memo about this company called Eve and compare it to every other company in the legal space that had ever done something before.'"

-- Alex Rampell

This quote illustrates the fundamental difference between raw data or model capabilities and a finished, valuable product. While PitchBook provides raw data, the real value is extracted by an analyst who synthesitses it into actionable insights. Similarly, AI models are the raw materials; the applications are the finished meals. Companies that can effectively transform these raw materials into indispensable tools, particularly by owning the end-to-end workflow, create significant defensibility. This is where the concept of "hostages, not customers" comes into play -- companies that embed themselves so deeply into a user's workflow that switching becomes prohibitively difficult.

Layer 1: The AI-Native Transformation of Existing Software

The first major investment theme is the AI-native transformation of traditional software categories. This isn't about incremental improvements; it's about fundamentally rebuilding existing software with AI at its core. The analogy of investing in "cloud-native" companies twenty years ago, which yielded incredible returns, is potent. Today, the equivalent is investing in AI-native software.

The danger for incumbents is their legacy. Companies like NetSuite or Salesforce, while powerful, may struggle to adapt as quickly or as comprehensively as new AI-native entrants. This is particularly true when new companies can offer AI-powered features that solve problems incumbents haven't even addressed. For example, Workday might offer AI-driven reference checks for $500, but an AI-native competitor could potentially offer a similar, or even superior, service for a fraction of the cost, or as part of a broader, more integrated solution.

The distinction between "greenfield" and "brownfield" opportunities is crucial here. Brownfield refers to competing in an existing market, like offering an "AI Mailchimp." This is challenging because users are already invested in existing solutions. Greenfield, however, involves targeting new companies or companies at an inflection point where they are forced to re-evaluate their software stack. For instance, a startup experiencing rapid growth and needing multi-entity, multi-currency support might be more open to adopting a new AI-native ERP system like Rillet over a legacy solution.

The implication for competitive advantage is that incumbents will likely integrate AI to enhance their existing offerings, making them stronger. However, they may be slower to innovate in entirely new ways or to disrupt their own established revenue streams. This creates an opening for AI-native companies to capture market share, especially in greenfield scenarios or by offering fundamentally different value propositions, like paying per outcome rather than per seat.

Layer 2: Software That Directly Replaces Labor

The second, and arguably most exciting, theme is the emergence of new software categories where the software directly replaces human labor. This is a vastly larger market than traditional software, as it taps into the fundamental human desire to be "richer and lazier."

The critical distinction here is that this isn't just about automation; it's about delivering value that was previously only achievable through human effort, often at a significantly lower cost. Consider a small business needing a receptionist. Hiring a human involves significant cost, training, and limitations (e.g., working hours, language barriers). An AI-powered software solution that can handle 90% of those tasks, even if not perfectly, becomes incredibly attractive.

The challenge and opportunity lie in pricing and defensibility. Companies can't simply charge the equivalent of a human salary. Instead, they must find a sweet spot that offers compelling value. Furthermore, simply replacing a human task isn't enough; the software needs to become a "system of record" to build a moat. This means owning the entire workflow, not just a single task. The example of Eve, an AI platform for plaintiff attorneys, is illustrative. By owning the end-to-end workflow from intake to case management, Eve not only makes attorneys more productive but also generates proprietary data that further strengthens its position.

"The predominant theme is that you have a lot of things where you would hire a person. You can't hire that person, or that person that you were going to hire doesn't speak 21 different foreign languages and won't work 24 hours a day. But software can do 90% of what that human would do."

-- David Haber

This highlights the core value proposition: AI can do what humans can't or won't, at a scale and cost that makes it economically viable. The risk for new entrants is that if they only offer a single, easily replicable AI function, they become vulnerable to underpricing. Building a sticky, end-to-end workflow is essential for long-term success. This is where companies like Salient, which collects 50% more revenue in auto loan servicing, demonstrate immense value generation beyond mere cost savings.

Layer 3: Walled Gardens of Proprietary Data

The third theme, "walled garden businesses," focuses on companies that build defensible moats around proprietary data. In an era where AI models are becoming increasingly commoditized, unique and proprietary data becomes the ultimate differentiator.

The key insight is that while raw data might be free or publicly available (like flight transponder data or legal records), the value lies in its aggregation, digitization, and the AI-powered insights derived from it. Companies like FlightAware, PitchBook, and Ancestry.com built their businesses by collecting and organizing vast datasets that were previously inaccessible or difficult to use. AI amplifies this value exponentially.

For example, OpenEvidence, by licensing exclusive access to medical journals, can provide AI-powered insights to doctors that ChatGPT, with its general knowledge, cannot replicate. Similarly, Vlex, by aggregating Spanish legal records, was able to quintuple its revenue by adding AI capabilities. The proprietary data itself, combined with AI, creates a finished product that is far more valuable than the raw components.

"The reason why I mention this is what do you do with FlightAware data? Or what do you do with Bloomberg data? Or what do you, like I'll tell you what I do with PitchBook data. I hire an analyst, and I say, 'Analyst, go write me a memo about this company called Eve and compare it to every other company in the legal space that had ever done something before.'"

-- Alex Rampell

This quote underscores the shift from selling raw data to selling finished, AI-enhanced products. The value accrues not just from having the data, but from the ability to process it in a way that is unique, inaccessible elsewhere, and directly solves a customer's problem more effectively than generic AI models. These walled gardens become incredibly sticky because they offer a curated, high-value experience that competitors cannot easily replicate, especially when the underlying data is not publicly available or easily accessible.

Key Action Items

• Prioritize Workflow Ownership: Focus on building AI solutions that manage entire end-to-end workflows rather than isolated tasks. This creates stickiness and defensibility. (Immediate Action)
• Identify Greenfield Opportunities: Seek out emerging markets or inflection points where new AI-native solutions can displace legacy systems or create entirely new categories, rather than competing in saturated brownfield markets. (Ongoing Strategy)
• Develop Data Moats: Actively seek out and secure unique, proprietary data sources. Invest in the infrastructure to collect, digitize, and enrich this data, then leverage AI to create unique, high-value products. (12-18 Month Investment)
• Embrace "Labor Replacement" Value: Explore opportunities where AI can perform tasks previously done by humans, focusing on delivering significant value and cost savings, and building systems of record around these capabilities. (Immediate Action)
• Re-evaluate Existing Software: For companies using traditional software, critically assess whether AI-native alternatives offer a fundamentally better value proposition, especially for new initiatives or at points of significant business change. (Ongoing Evaluation)
• Build for Outcome-Based Pricing: Shift from per-seat or per-feature pricing to models that reflect the value and outcomes delivered by AI-powered solutions, particularly in labor-replacement categories. (Immediate Action)
• Invest in AI-Native Talent: Ensure your team possesses the expertise to develop and deploy AI-native applications, focusing on product development that integrates AI seamlessly into user workflows. (Ongoing Investment)