I’ve covered the tech industry for decades, and over that time, I have witnessed all the big platform shifts - from mainframe to PC, from PC to the internet and from the internet to mobile. Each time, a moment came when the existing software and infrastructure became too constraining, forcing everything to be rebuilt to accommodate the new paradigm. We are at this juncture once again, but this time it is being pushed by AI agents.

While the leading companies have been busy over the past few years programming agents to use software in the manner of a typical person (clicking buttons, filling out forms), the top players have decided to build the software for agents. This is more significant than most understand.

Building for Agents, Not Humans

Imagine agents as self-driving cars operating on infrastructure built for human drivers, with stoplights, painted lanes and other rules designed for people. That creates obvious inefficiencies. The smarter approach is to redesign the infrastructure to match what agents do best.

Yann LeCun, former head of Meta AI and a luminary in this space, succinctly described the approach: Agents will communicate effectively with each other without pretending to click buttons. That is not a hypothesis but rather a technical inevitability.

The New Agentic Infrastructure Layer

Anthropic, among others, foresaw the future and created the Model Context Protocol in advance, offering developers a standardized protocol for connecting AI systems to various tools and databases. That is infrastructure thinking, which plays a significant role in the matter.

Other examples include Stripe, Mastercard and OpenAI creating payment rail infrastructure to enable agents to execute the task of agentic shopping. Salesforce launched the Headless 360 platform , allowing enterprises to make their software directly accessible to agents without requiring any UIs.

The company that best exemplifies this change is Zapier . The firm managed to take chaotic software interfaces and distill them into clear, programmable actions. It turns out, this architecture suits agents well. The Zapier CEO, Wade Foster, explained to Axios that his team expects AI agents to emerge as dominant consumers of software. Instead of competing with humans, agents will complement them, significantly increasing the number of transactions.

Aaron Levie, CEO of Box and one of the most thoughtful leaders in the enterprise tech space, succinctly identified the implications: "If agents are consuming your service or products, you should not expect agents to consume it through a graphical interface. They will consume the product through the API."

What this means is that competitive advantage in the age of agents will not come to those who develop great-looking software UIs, but rather to companies that control APIs, data flows, and access controls.

When mobile computing took over, the winners were not the firms trying to squeeze their bulky software into small screens but rather startups developing solutions from the ground up, for smartphones. And now it will be the same again.

As usual, eventually, hardware will catch up with the software shift. Already, there are experiments such as AI-first devices, always-on microphones built into eyewear and jewelry and a long-awaited Apple-like device from OpenAI, designed by Jony Ive. However, this process is much more difficult to implement than software changes. With every new generation of hardware, costs rise exponentially as development cycles lengthen and distribution channels become less efficient.

I would expect a software lead over hardware innovation for several years, until the new use cases are established and developed enough to create value.

We are witnessing the start of fundamental re-platforming of both enterprise and consumer software stacks. Companies that understand this early enough and begin building their agent-optimized infrastructure will drive the next wave of innovation. Otherwise, they will remain as irrelevant as those who missed the boat of building mobile-first solutions.

Each platform transition I’ve covered has rewarded the companies that saw the shift early and acted decisively. This one will be no different.