Adobe Exec Predicts Commoditization of AI Models, Urges Focus on Application Layer

As AI continues to rapidly evolve, the underlying large language models (LLMs) powering many of today's applications are headed for commoditization, putting pressure on model companies to move up the stack and capture value in higher-level applications and tools, according to Scott Belsky, Chief Strategy Officer at Adobe.

The Inevitable Tide of Commoditization

Belsky's warning comes amid a growing consensus in the industry that the value is shifting from foundational models to the applications and services built on top of them. This shift is driven by factors such as decreasing costs, the rise of open-source alternatives, and the increasing importance of specialized knowledge and real-world application.

"I think that most models will be commoditized," Belsky said in a recent interview with a16z. "Chat GPT 3.5 is now 98% cheaper than it was when it launched and 4.0 is already 90% cheaper. It's just so dizzying."

Belsky highlighted the risk of being a "wrapper app" that relies solely on a commoditized model and faces the threat of being replaced by a platform feature. He used the analogy of the flashlight app, which was a profitable application until Apple shipped the flashlight as a native feature of its operating system.

This viewpoint echoes the sentiment shared by Infosys chair Nandan Nilekani, who stated last year that "the models will become more commoditized and the value will switch to the application layer and the whole stack," as well as Microsoft CEO Satya Nadella, who recently tweeted that AI is going to become a commodity "we just can't get enough of."

AI Agents: A Catalyst for Change

The emergence of AI agents, which can perform complex tasks autonomously, is further accelerating the commoditization of LLMs. These agents rely on LLMs, but their value lies in their ability to integrate with data, automate workflows, and provide intelligent services, not simply in the underlying model itself.

As these agents become more sophisticated and accessible, the underlying LLMs become less of a differentiating factor, further pushing the industry towards commoditization.

Moving Up the Stack

Recognizing this shift, Adobe is adapting its strategy to focus on higher-value applications and services. Belsky emphasized the company's decision to become a platform-agnostic model company, supporting all major AI models in addition to building its own.

"One of our major decisions that we've made at Adobe over the last year is to actually become a platform agnostic model company," Belsky stated. "So we do build our own models in a way that's commercially safe and is important to a lot of our customers, but we're also going to support all the major models and people are going to be able to choose which model they want for which action."

The focus of AI is now shifting to systems that are combining language models and contextual business data. Belsky emphasizes that if you've raised billions of dollars to build these fundamental models that are ultimately commoditized, you absolutely have to move up the stack. Much like the success of AWS with their move up the stack, Belsky believes that this is the way to capture the margin of the business.

"These model companies that are currently powering all these products are going to have to go up the stack to capture the value in the business," says Belsky.