Is the Application Layer Dead?

Four reasons AI applications still have a right to win

Over the past year, the argument around the AI application layer has shifted.

As foundation models have become more powerful, model companies have begun expanding beyond the intelligence layer and into applications for specific industries and functions. Anthropic, for example, has introduced capabilities for areas like legal work and finance. That has led some people to ask whether companies building specialized AI applications still have a future.

If a general-purpose model can review an NDA, help build a financial model, or complete other professional tasks, do we still need a legal AI company, a finance AI company, or any other vertical application?

It is a fair question. But I think the argument that foundation models will absorb the entire application layer is incomplete.

Model companies are building incredibly valuable infrastructure. They may also attempt to move further up and down the stack, particularly given the scale they need to reach to justify their valuations. But that does not mean they automatically have the right to win every layer, every industry, and every workflow.

There are at least four reasons specialized application companies will continue to matter.

1. Real-world work is more nuanced than a general-purpose tool can easily capture

Foundation models are extremely powerful, but most professional work is not a collection of simple, interchangeable tasks.

Legal, finance/accounting, healthcare, and other professional functions each contain their own terminology, processes, judgment calls, and organizational context. The deeper you go into one of these industries, the more complex the workflow becomes.

A general-purpose model may be perfectly capable of helping an individual review a straightforward NDA for example. But that is very different from deploying a legal AI system across a large law firm or an enterprise legal department that can cover different use cases (e.g. litigation, due diligence, etc.).

Most people are also busy doing their actual jobs. They are not AI experts, and they do not necessarily know how to take a broad horizontal tool and configure it for a highly specific professional workflow. Foundation model companies have a great deal of wood to chop simply getting more people and organizations to adopt their core products. It does not necessarily make sense for them to master every complex workflow within every individual industry.

The argument that one model company can become the end-all, be-all for every application downplays how nuanced most professional work really is.

2. Regulation and compliance require vertical expertise

The need for specialization becomes even clearer in regulated industries.

Healthcare is an obvious example. A company selling into healthcare may need to account for privacy requirements, compliance standards, clinical workflows, organizational policies, and jurisdiction-specific regulations. Those requirements may also change depending on the type of customer and the geography in which the product is being deployed.

For a foundation model company to serve that industry comprehensively, it would need to hire a specialized workforce, build a product roadmap around the vertical, understand the relevant regulatory bodies, and continually adapt the product to the way healthcare organizations operate.

That is a significant amount of work for one market. The same is true in legal services, financial services, insurance, and many other industries.

A vertical application company, by contrast, can make those requirements central to the product from the beginning. It can organize and store information in a way that fits the industry, design around the customer’s regulatory environment, and build trust with users who require a very high degree of reliability. The more consequential the work, the more important that expertise becomes.

A general-purpose tool may be able to perform certain tasks within a regulated industry. But performing an isolated task is not the same as delivering an enterprise-ready product that an organization can safely and consistently adopt.

3. Enterprises need measurable ROI, not just access to AI

One of the most common questions we hear from enterprises today is not, “How do we get access to AI?” 

Most organizations already have access to it. The harder questions are: Where should we deploy it? Who should use it? How do we know whether it is working? And how do we measure the return on our investment?

A company may know that employees throughout the organization are using Claude, ChatGPT, or another general-purpose model. But it may not have a clear view into who is using it, how they are using it, or whether that usage is producing meaningful value.

Vertical applications give companies a clearer unit of adoption and ROI. An organization can deploy one product within its legal team, another within sales, and another within engineering. It can then evaluate whether each team is completing work faster, producing better outcomes, reducing costs, or taking on more complex responsibilities.

The application provider also has a direct incentive to understand the end user. It can build its roadmap around the work that person performs every day, observe where the product is creating value, and continue expanding into more complex parts of the workflow. That often requires a human element.

Many leading application companies are hiring forward-deployed engineers or industry experts who can sit alongside customers, train employees, understand how the organization works, and help embed the technology into existing processes. 

Vertical AI companies are not simply giving customers a tool and asking them to figure it out. They can help the organization understand where the product belongs, how to use it, and how to achieve measurable value from it.

It would be difficult for a horizontal platform to offer that depth of implementation across every possible industry. Nor should that necessarily be its focus.

As boards and leadership teams begin asking more rigorous questions about AI spending, the ability to demonstrate clear ROI will become increasingly important. Application companies are often in a much better position to provide that answer.

4. Customers want accountability and an enterprise-ready product

There is also an accountability question. When a company purchases a dedicated application for an important workflow, it is buying a product that is intended to deliver a specific outcome. There may be service-level agreements, uptime commitments, performance expectations, and a vendor responsible for helping resolve issues when something goes wrong. That creates a much clearer relationship.

The customer can say, “You sold us a product to perform this function, and it is not delivering what we agreed upon.” The vendor can then troubleshoot the issue, improve the product, and work with the customer to reach the intended outcome.

That is very different from taking a horizontal or open-source tool and adapting it internally. In that situation, the organization may need its own technical experts to configure the product, maintain it, troubleshoot it, and determine why it is not working as intended. Many companies do not have the talent, resources, or time to do that. They need something that is ready to implement and supported by a team that understands the problem they are trying to solve.

A plugin or general-purpose model may provide impressive underlying technology. But, in some ways, it is similar to an open-source solution: the technology may be strong, while significant work is still required to make it usable for a specific organization and workflow.

For many enterprises, it will be much easier to purchase a purpose-built solution that works out of the box, comes with support, and has a vendor accountable for its performance.

The infrastructure is only the beginning

The current AI buildout has parallels to the early internet era. Infrastructure companies captured enormous value as the internet was being built, and that value never went away, companies like AWS remain foundational and highly valuable today. But a second, larger wave followed: the SaaS era, when the products people and businesses used every day were built on top of that infrastructure.

We may still be in the early innings of a similar shift in AI.

Today, much of the value is accruing to the model and infrastructure layers, and the scale of their growth is extraordinary. But as the infrastructure matures, the question will increasingly become what people build on top of it and how those products improve real work.

The application layer is not dead. We are just beginning to see what it can become.