AI as Universal Interface: How Agents Translate Intent Into Software

Key Insights

• Software has its own grammar: Every tool—Figma, Salesforce, CAD—requires learning a specialized vocabulary (frames, accounts, splines) that rises in complexity with capability.
• AI speaks English: Rather than forcing users to learn software's native language, AI agents translate intent expressed in plain English into the application's underlying logic.
• The expertise gap closes differently: Instead of traditional UX design, AI-powered products shift focus to documentation and agent access to system schemas, letting users skip the learning curve.
• Expertise doesn't disappear—it transforms: Like software engineers moving from writing syntax to architecting systems, domain experts now focus on what agents need to operate safely and completely.

The Grammar Problem in Software

Every powerful software product has its own language. Figma speaks in layers and components, Salesforce in accounts and opportunities, CAD software in splines and constraints. Each demands a learning curve, and the relationship is unidirectional: the more powerful the tool, the steeper the expertise required.

This creates a fundamental onboarding burden for product-led growth companies that must sell themselves without a sales team. Traditional UX labs can close this gap one user at a time, but it's a slow, manual process.

English as a Universal API

AI inverts this problem by making English the interface. Instead of learning software's grammar, users simply describe their desired outcome in plain language. An AI agent then translates that intent into the application's native vocabulary.

This approach first succeeded in programming: AI tools like Codex translated English into C++, Rust, and TypeScript. Now the same pattern applies to application-layer systems. A founder wanting to design a dress inspired by Ruth Asawa's wire sculptures could ask an AI agent to operate CAD software on her behalf—software she didn't know how to use. The result was a previously unmanufacturable garment, created without traditional technical expertise.

If AI can translate intent into CAD's grammar, it can do the same for any software.

The Shift in Product Design

This change fundamentally redefines what product design means. Software built for AI agents must be complex and well-documented—the opposite of traditional design philosophy. Users prefer simplicity and intuitive interfaces, but agents need access to complete system schemas and the ability to reason over broader capabilities at once.

The product manager's job evolves from designing every screen to building the harness that lets agents act safely and completely. This makes documentation—traditionally the domain of infrastructure and developers—central to application design. Documentation now directly impacts how well an agent can operate the software on behalf of users.

The Future of Expertise

The expert doesn't disappear; expertise migrates upward. Software engineering has already experienced this shift: developers no longer focus on individual component configuration but on systems architecture—how pieces fit together. The same evolution happens across tools when agents become the primary operators.

This creates space for what product design legend Dieter Rams called earned simplicity: users say what matters in English, and the agent absorbs the complexity underneath. The question for product teams becomes: What would great design look like when agents, not human interfaces, are the primary user?

Conclusion

AI closes the expertise gap not by simplifying software, but by letting agents handle its complexity while users work in plain English. This shifts focus from onboarding individual users to enabling agents to operate systems completely and safely. As AI becomes the intermediary between human intent and software grammar, the winners will be products with clear, comprehensive documentation—and teams that understand the systems underneath.

---

Original source: Previously Unmanufacturable

powered by osmu.app