Most software has been designed around a simple assumption: the user is human.
Based on this assumption, digital products have been built around the needs and limitations of people, i.e., our attention spans, visual perception, and tolerance for friction. Entire disciplines emerged to refine this interaction. User experience design, growth optimization, onboarding flows, and branding all evolved to reduce friction for human beings navigating digital systems.
Still today, most software operates under this premise.
But that premise is beginning to change.
As AI agents become more capable, an increasing number of digital tasks are being delegated rather than directly performed. Already, agents are beginning to use computers and are writing code, reviewing documents, transacting across platforms, etc. I believe this will lead to a future where agents become the majority of software and internet users.
In this context, the effective consumer of software will no longer be humans, but the system acting on their behalf. This represents a shift in who the functional users of digital products are and is not merely a productivity improvement.
Designing for Agents
Today, software products differentiate themselves through clarity, aesthetics, usability, and brand perception. These factors matter because humans are influenced by presentation and emotion as much as by utility.
Agents are not.
An agent does not experience visual hierarchy or persuasive copy; it evaluates structured information like price, response time, compatibility, and output quality, making it care solely about the product's utility. This is a structural distinction, not just a technical one.
To survive, software will need to offer direct access to products and services through interfaces such as APIs, CLIs, MCP, or other protocols, so agents can interact with them natively. Without programmatic access, agents are forced to imitate human behavior by navigating websites and parsing layouts, effectively reducing the graphical interface to a fragile and inefficient API.
Agents will favor software with native integrations because it is more accessible, more reliable, and less error-prone. In an ecosystem where agents represent a growing share of activity, infrastructure becomes the primary interface.
The central design question will shift from, "How intuitive is this for a person?" to, "How efficiently can this be used by an agent?"
How Competition Changes
As agents begin choosing between products, competition becomes more quantitative and less perceptual. Small differences in price, latency, reliability, and integration depth become easier to compare and more consequential in shaping demand.
In this environment, machine legibility becomes a competitive advantage. Clear documentation, stable APIs, consistent performance, and native integrations matter more because they make a product easier for agents to discover, evaluate, and use. One could describe this as a shift from optimizing for human visibility to optimizing for machine legibility, or from Search Engine Optimization (SEO) toward Agentic Optimization (AO). As a result, optimization moves away from human-facing presentation and toward structured accessibility. Access becomes persuasion. Performance becomes usability.
This will change the mechanics of competition.
Interface vs. Infrastructure
The most defensible businesses may be those whose value lies beneath the interface layer.
When Dara Khosrowshahi, CEO of Uber, spoke at USC last year, I asked whether he could envision a future where Uber no longer needs a consumer-facing mobile app, and instead exposes APIs for personal AI assistants to request rides.
His response made it clear that this future is very plausible and noted that he could see Uber offering APIs to agents "all over the place," and that the Uber app itself could eventually be bypassed by third-party agents plugging directly into its network.
What he emphasized, however, was that Uber's real value does not sit in the interface. It sits in the marketplace, the proprietary supply network, and the algorithmic ability to price it better than anyone else. If agents become the new interface layer, the companies that remain defensible won't necessarily be those with the best apps, but those controlling the underlying infrastructure and supply. As he pointed out, businesses that merely aggregate third-party supply, like online travel agencies, may find themselves easily disintermediated in an agent-driven world.
This distinction is critical.
As demand becomes increasingly mediated by agents rather than by individuals tapping screens, the underlying system retains its value. Only the surface through which it is accessed changes.
The interface is an access layer. The infrastructure is the moat.
Companies that mistake interface polish for structural advantage will find themselves deeply vulnerable in a world where execution is delegated.
Early Signals
We are already seeing early indicators of this transition.
Agent harnesses like OpenClaw allow systems to interact directly with tools and services through structured, programmatic interfaces. Google has already released a native CLI tool for parts of its product suite, allowing agents to bypass the GUI entirely.
Products that provide stable, well-documented, and performant endpoints are already structurally better positioned for an agent-mediated environment.
Interfaces Won't Disappear
This does not imply that graphical interfaces will vanish entirely.
Humans will still require visibility, control, and trust. Interfaces may increasingly function as supervisory layers, such as dashboards, through which individuals set objectives, review outputs, and override decisions, while agents execute.
As delegation expands, the execution layer of the internet will increasingly be operated by machines.
Conclusion
While this transition is still in its early stages, the broader directional shift is undeniable. Standards for agent integrations have not yet been set in stone, and we are already seeing protocols like MCP beginning to be questioned. But the underlying trajectory remains constant. Agents will increasingly use software on our behalf, and over time, they will become the majority users of software and the internet. To survive this shift, companies must stop optimizing solely for human attention and start building for the customers of the future.