Agentic Ergonomics and the Barcode Web
The era of human-readable code is over as software architecture shifts to verbose, deterministic frameworks optimized entirely for AI agents.
By Theo Marsh
Sparked by The front end framework for correctness: built on Effect, architected like Elm · discussion
There was a spicy thread on Hacker News this weekend reacting to Foldkit, a new UI framework. The consensus was mostly horror. Developers scrolled through the documentation and immediately recoiled at how boilerplate-heavy and verbose the syntax is. They complained about the sheer volume of code required to render a simple list view, comparing the experience to writing assembly language for the web.
The critics are judging it on the wrong axis.
To see why, watch what happens when engineers push Claude to generate complex UI components in modern React. React hooks are brilliant for human developers because they hide complexity behind elegant abstractions. We love implicit state. We love clean dependency arrays that magically track our variables. But for a language model, implicit context is a massive footgun. If you monitor an LLM trying to maintain a complex React application over multiple turns, you watch it confidently hallucinate stale closures. It will write a perfectly formatted useEffect hook that captures an old version of a variable, completely breaking the application state. The framework relies on invisible connections that humans intuitively grasp and machines routinely drop.
Elegant abstractions force the machine to guess.
In 1973, the grocery industry faced a similar ergonomic crisis. For decades, stores had stamped human-readable prices like "$1.99" onto individual packs of Wrigley's Juicy Fruit gum. It was incredibly friendly for the cashiers, who could read the sticker & punch the number into the register, and friendly for the shoppers. Then the industry replaced those friendly numbers with the Universal Product Code.
The human cashiers hated it. The new tags were ugly, high-contrast squares of black and white that nobody could read by eye. You couldn't glance at a pack of gum and know what it cost anymore.
But the IBM 3660 scanner didn't care about human elegance. It wanted deterministic, explicit data. The laser hardware demanded a new, non-human syntax. Software architecture is about to undergo the exact same transition.
We are entering the era of Agentic Ergonomics.
We can completely ignore Foldkit's standard feature list here. Comparing its rendering speed or bundle size to React completely misses the broader tectonic shift. We are no longer the primary authors of code. For the last fifteen years, the industry optimized frameworks for human minds. We praised DRY principles. We built syntactic sugar to keep files small and visually clean so a developer could hold the entire component logic in their working memory.
Agentic Ergonomics optimizes for the exact opposite. It favors verbose, deterministic architectures that explicitly pass every piece of state down the chain. It looks exhausting to a human eye. It is full of rigid repetition. But that repetition acts as a perfectly scannable barcode for an LLM. When every dependency is explicitly declared and nothing is left to a clever underlying engine, an agent can confidently parse, modify, and rewrite the file without accidentally severing an invisible state wire.
The machine wants the barcode.
You might read this and immediately raise the obvious objection: humans still have to debug this code when the agent fails. If a framework generates ten thousand lines of boilerplate for a basic dashboard, a human engineer will drown trying to find the bug.
My guess?
They won't even try.
Or rather, they will just use an agent to read the barcode back to them. We are optimizing for the entity actually doing the typing. When you look at a modern fulfillment warehouse, the shelves are no longer organized by product category so a human picker can find the shoes next to the socks. They are organized via chaotic, algorithmic binning that makes zero sense to a person but allows a robotic retrieval system to operate at maximum efficiency. Codebases are heading toward that same algorithmic binning. The human won't read the code.. they will ask the system why the button is red, and the system will read its own barcode to find out.
The resistance to verbose frameworks right now is just the friction of cashiers realizing they can no longer read the price tags on the gum. They are mourning the loss of a human-centric system. And they are right to mourn it, because the shift is irreversible.
The fifteen-year arms race for human-readable front-end frameworks is over. By the early 2030s, the most popular web framework in the world won't be written by humans, and it won't be readable by them either. It will just be a barcode. And it will work.