Previously Unmanufacturable

In short : Every software product has its own grammar: Figma has frames and components, Salesforce has accounts, contacts, and opportunities, CAD has tools and constraints. Capability and required expertise rise together, so the most powerful tools are the ones fewest people can operate. AI lets users express intent in English while an agent handles the application's grammar, illustrated by a non-technical founder who directed Codex to drive CAD software and produce a dress once previously unmanufacturable for her. The counterpoint is that experts do not disappear: as in software engineering, where agents write most of the code but hard problems still require someone who understands the system underneath, depth still matters in the systems where depth is the job.

“This garment is previously unmanufacturable.”

How I AI, on Yana Welinder¹

Software is powerful only after someone learns its interface. Figma speaks in layers & components, Salesforce in accounts & opportunities, CAD in splines & constraints. Each vocabulary takes time to learn, & the relationship runs in one direction : the more powerful the software, the more expertise it demands, regardless of what a product manager intends. A usability lab exists to close that gap, one user at a time.²

AI closes it a different way, by making English the interface.

That onboarding problem burdens product-led growth companies, the ones built to sell themselves without a sales team, as much as forward-deployed engineering products. English is becoming a universal API : a person describes the outcome, & an agent translates that intent into computerese.³ Programmers first spoke this way through AI’s babelfish : english to C++, Rust, & TypeScript. Increasingly, the languages worth translating are those of the application layer : accounts, purchase orders, circles, & splines.

Inspired by Ruth Asawa’s looped-wire sculptures, Yana Welinder wanted to create a previously unmanufacturable dress. So, she asked AI to use software she couldn’t operate on her behalf.¹ The result is spectacular.

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

This inverts product design. Software built for AI is complex & well-documented. People prefer simplicity, never-read-the-manual, let me jump in & try first.⁴

An agent can hold, retrieve, & reason over more of a system’s schema at once than a person clicking through menus, so the product manager’s job shifts from designing every screen to building the harness that lets an agent act on the system safely, completely, & without needing to be shown where anything is.

Documentation matters more in this world. Infrastructure has always depended on it, because developers need to understand APIs, schemas, & failure modes before they trust a system. That requirement is moving up the stack, into applications.⁵ Designing that documentation will be an essential part of product design & the UX lab evolves into evaluations - watching how an agent operates software.

Nor does the expert disappear. Software engineering’s renaissance has replaced syntax editing with systems architecture : how the pieces fit together, not configuring individual components.

Welinder’s dress demonstrates the higher level of abstraction AI affords. To enable this end-user simplicity, new software must expose greater complexity to agents. Dieter Rams, arguably the greatest product designer ever, started from the premise that everything is noise, then earned the right to keep only what was essential.⁶ English lets a person do the same with software : say what matters, & let the agent absorb the rest. What would Rams design in this era?

Get the next one in your inbox

The 1-minute read that turns tech data into strategic advantage.
Read by 150k+ founders & operators.

GP at Theory Ventures. Former Google PM. Sharing data-driven insights on AI, web3, & venture capital.

Bloomberg • WSJ • Economist