The Future Is Agents

Software is shaped by its users. And since the beginning, the users have been human. Interfaces, workflows, and business models have all been built around human attention and behavior. The question is how long that remains true – whether human priorities stay primary as software systems become more autonomous.

Early cracks in this assumption appeared with search. The web began to optimize not just for people, but for ranking algorithms. SEO turned content into something written for machines first, humans second. Today, the web serves two users – human and algorithmic – and the balance has shifted noticeably toward the latter (have you seen the web recently?).

App stores extended this dynamic. Mobile software came to depend on opaque ranking systems and review guidelines. Visibility – and therefore economic survival – was shaped less by user preference than by compliance with gatekeeping systems.¹ Software was still used by humans, but increasingly designed for intermediaries.

Now a different kind of user is emerging – one that, I believe, does not just guide human action but replaces it. These systems act on behalf of users: executing tasks, navigating interfaces, and producing outcomes with minimal supervision. This is not a change in interface. It is a change in the user itself.

Call this agent-centric design.

An agent, in this sense, is not a chatbot or a recommendation system. It is software that consumes other software directly – making decisions, invoking tools, and completing multi-step tasks without continuous human input or oversight. The human defines the goal, however abstract; the agent executes. Booking travel, reconciling expenses, drafting communications; these become machine-to-machine interactions, reviewed only after the fact.

This shift is no longer theoretical. New protocol layers are being built explicitly for agents. Anthropic’s Model Context Protocol² enables structured communication between agents and tools. Browser-native variants are emerging.³ Major platform providers are integrating these capabilities across core products.⁴ Google and Microsoft are tilting their software from humans to agents before our eyes.

I want to preempt three objections:

One: this is just chatbot hype. It is not. Chatbots are interfaces; protocols are infrastructure. The distinction matters. Interfaces shape experience; protocols define participation. When the protocol layer changes, the system reorganizes around it.

Two: humans will remain in the loop. They will, but at a different level. Approval replaces interaction. The primary unit of consumption shifts from human attention to agent execution. Interfaces become fallbacks; APIs become the product.

Three: the shift will be slow. Perhaps. But design assumptions lag reality. Systems being built today under a human-only model are already misaligned with where platforms are heading.

When the user changes, everything downstream changes with it.

Software built for humans optimizes for attention: layout, persuasion, navigation. Software built for agents optimizes for execution: clarity, structure, and completeness of action. Humans read screens; agents read interfaces. Humans browse; agents act.

This breaks existing models. Advertising depends on human attention. Interface friction can be monetized when a person is clicking – but not when an agent is completing tasks directly. When agents become the primary operators, the economics of software shift with them.

The companies that treat agents as secondary users will optimize for interpretation – hoping their software is understood. The companies that treat agents as primary will optimize for execution – ensuring their software is usable by machines. One is legible. The other is operable.

That difference compounds.

Agents are not a feature. They are a new class of user, and I believe they are about to declare themselves dominant. I’ve extended this argument – that the substrate itself is shifting, and AI-native apps will swallow the web – in a follow-up.

The remaining question is whether software is designed for them deliberately – or continues to evolve in their direction by accident.