Midjourney engineer debuts new vibe coded, open supply customary Pretext to revolutionize internet design

For 3 a long time, the internet has existed in a state of architectural denial. It is a platform initially conceived to share static physics papers, but it is now tasked with rendering the most complicated, interactive, and generative interfaces humanity has ever conceived.

At the coronary heart of this stress lies a single, invisible, and prohibitively costly operation referred to as “layout reflow.” Every time a developer wants to know the top of a paragraph or the place of a line to construct a contemporary interface, they have to ask the browser’s Document Object Model (DOM), the customary by which builders can create and modify webpages.

In response, the browser usually has to recalculate the geometry of the whole web page — a course of akin to a metropolis being compelled to redraw its whole map each time a resident opens their entrance door.

Final Friday, March 27, 2026, Cheng Lou — a distinguished software program engineer whose work on React, ReScript, and Midjourney has outlined a lot of the trendy frontend panorama — announced on the social network X that he had “crawled via depths of hell” to launch an open source (MIT License) solution: Pretext, which he coded utilizing AI vibe coding instruments and fashions like OpenAI’s Codex and Anthropic’s Claude.

It is a 15KB, zero-dependency TypeScript library that enables for multiline textual content measurement and format fully in “userland,” bypassing the DOM and its efficiency bottlenecks.

With out getting too technical, in brief, Lou’s Pretext turns textual content blocks on the internet into absolutely dynamic, interactive and responsive areas, ready to adapt and easily transfer round some other object on a webpage, preserving letter order and areas between phrases and features, even when a consumer clicks and drags different objects to intersect with the textual content, or resizes their browser window dramatically.

Paradoxically, it is troublesome with mere textual content alone to convey how important Lou’s newest launch is for the whole internet going ahead. Fortuitously, different third-party builders whipped up fast demos with Pretext displaying off a few of its extra spectacular powers, together with dragon that flies around within a block of text, respiration hearth as the surrounding characters soften and are pushed out of the means from the dragon’s undulating type.

One other guy made an app that requires the consumer to hold their smartphone precisely degree, horizontal to learn the textual content — tipping the gadget to one aspect or the different causes all the letters to fall off and gather there as if they had been every bodily objects dumped off the floor of a flat tray. Somebody even coded an online app permitting you to watch a whole movie (the new Project Hail Mary starring Ryan Gosling) whereas studying the ebook it is primarily based on at the identical time, all rendered out of interactive, shifting, quick, responsive textual content.

Whereas some detractors instantly identified that many of those flashy demos make the underlying textual content unreadable or illegible, they’re lacking the bigger level: with Pretext, one man (Lou) utilizing AI vibe coding instruments has singlehandedly revolutionized what’s doable for everybody and anybody to do when it comes to internet design and interactivity. The challenge hasn’t even been out every week — after all the preliminary customers are solely scratching the floor of the newfound capabilities which heretofore required complicated, customized directions and will not be scaled or generalized.

After all, designers and typographers could also be the ones most instantly impressed and affected by the advance — however actually, anybody who has frolicked attempting to lay out a block of textual content and wrap it round pictures or different embedded, interactive parts on a webpage is most likely going to have an interest on this. However anybody who makes use of the internet — all 6 billion and counting of us — will probably expertise a few of the results of this launch before too lengthy because it spreads to the websites we go to and use day by day.

And already, some builders are working on more useful features with it, like a customized user-controlled font resizer and letter spacing optimizer for these with dyslexia:

With that in thoughts, maybe it is not suprising to be taught that inside 48 hours, the challenge garnered over 14,000 GitHub stars and 19 million views on X, signaling what many consider to be a foundational shift in how we construct the web.

It additionally demonstrates that AI-assisted coding has moved past producing boilerplate to delivering basic architectural breakthroughs. For enterprises, this signifies a brand new period the place high-leverage engineering groups can use AI to construct bespoke, high-performance infrastructure that bypasses decades-old platform constraints, successfully decoupling product innovation from the gradual cycle of industry-wide browser standardization

The geometry of the bottleneck

To know why Pretext issues, one should perceive the excessive value of “measuring” issues on the internet. Customary browser APIs like getBoundingClientRect or offsetHeight are infamous for triggering format thrashing.

In a contemporary interface—consider a masonry grid of 1000’s of textual content containers or a responsive editorial unfold—these measurements occur in the “scorching path” of rendering. If the browser has to cease and calculate format each time the consumer scrolls or an AI generates a brand new sentence, the body price drops, the battery drains, and the expertise stutters.

Lou’s perception with Pretext was to decouple textual content format from the DOM fully. Through the use of the browser’s Canvas font metrics engine as a “floor fact” and mixing it with pure arithmetic, Pretext can predict precisely the place each character, phrase, and line will fall with out ever touching a DOM node.

The efficiency delta is staggering. In accordance to challenge benchmarks, Pretext’s format() perform can course of a batch of 500 completely different texts in roughly 0.09ms. In contrast to conventional DOM reads, this represents a 300–600x efficiency improve. This pace transforms format from a heavy, asynchronous chore right into a synchronous, predictable primitive—one that may run at 120fps even on cellular units.

Expertise: the put together and format cut up

The class of Pretext lies in its two-stage execution mannequin, designed to maximize effectivity:

• put together(textual content, font): This is the one-time “heavy lifting” section. The library normalizes whitespace, segments the textual content, applies language-specific glue guidelines, and measures segments utilizing the canvas. This end result is cached as an opaque knowledge construction.

• format(preparedData, maxWidth, lineHeight): This is the “scorching path”. It is pure arithmetic that takes the ready knowledge and calculates heights or line counts primarily based on a given width.

As a result of format() is simply math, it may be referred to as repeatedly throughout a window resize or a physics simulation with none efficiency penalty. It helps complicated typographic wants that had been beforehand not possible to deal with effectively in userland:

• Combined-bidirectional (bidi) textual content: Dealing with English, Arabic, and Korean in the identical sentence with out breaking format.

• Grapheme-aware breaking: Making certain that emojis or complicated character clusters are not cut up throughout strains.

• Whitespace management: Preserving tabs and exhausting breaks for code or poetry utilizing white-space: pre-wrap logic.

The hell crawl and the ai suggestions loop

The technical problem of Pretext wasn’t simply writing the math; it was guaranteeing that the math matched the “floor fact” of how varied browsers (Chrome, Safari, Firefox) really render textual content. Textual content rendering is notoriously riddled with quirks, from how completely different engines deal with kerning to the specifics of line-breaking heuristics.

Lou revealed that the library was constructed utilizing an “AI-friendly iteration methodology”. By iteratively prompting fashions like Claude and Codex to reconcile TypeScript format logic towards precise browser rendering on huge corpora—together with the full textual content of The Nice Gatsby and numerous multilingual datasets—he was ready to obtain pixel-perfect accuracy with out the want for heavy WebAssembly (WASM) binaries or font-parsing libraries.

Ripple results: a weekend of demos

The discharge of Pretext instantly manifested as a collection of radical experiments throughout X and the broader developer neighborhood. The original demos showcased by Lou on X offered a glimpse into a brand new world:

• The editorial engine: A multi-column journal format the place textual content flows round draggable orbs, reflowing in real-time at 60fps.

• Masonry virtualization: A demo displaying tons of of 1000’s of variable-height textual content containers. Top prediction is decreased to a linear traversal of cached heights.

• Shrinkwrapped bubbles: Chat bubbles that calculate the tightest doable width for multiline textual content, eliminating wasted space.

The neighborhood response was equally explosive. Inside 72 hours, builders started pushing the boundaries:

• @@yiningkarlli carried out the Knuth-Plass paragraph justification algorithm, bringing high-end print typography—decreasing “rivers” of white house by evaluating whole paragraphs as models—to the internet.

• @Talsiach constructed “X Times,” an AI-powered newspaper that makes use of Grok to analyze pictures and X posts, utilizing Pretext to immediately format a front-page reflow.

• @Kaygeeartworks demonstrated a Three.js fluid simulation that includes fish swimming via and round textual content parts, with the textual content reacting to physics at excessive body charges.

• @KageNoCoder launched Pretext-Flow, a stay playground for flowing textual content round customized media like clear PNGs or movies.

• @cocktailpeanut and @stevibe demonstrated ASCII art Snake and Hooke’s Law physics with stay textual content reflow.

• @kho constructed a BioMap visualization with 52 biomarker blocks performing format reflow at 0.04ms each body.

Philosophical shifts and the thicker shopper

The response to Pretext was overwhelmingly enthusiastic from frontend luminaries. Guillermo Rauch, CEO of Vercel, and Ryan Florence of Remix praised the library’s efficiency beneficial properties. Tay Zonday famous the potential for neurodiverse high-speed studying via dynamic textual content rasterization.

Nevertheless, the launch additionally ignited a nuanced debate about the way forward for internet requirements. Critics warned of “thick shopper” overreach, arguing that bypassing the DOM strikes us away from the simplicity of hypermedia methods. Lou’s response was a meditation on the lineage of computing. He pointed to the evolution of iOS—which began with PostScript, a static format for printers, and developed into a elegant, scriptable platform. The net, Lou argues, has remained caught in a “doc format” mindset, layering scripting on prime of a static core till complexity reached some extent of diminishing returns. Pretext is an try to restart that dialog, treating format as an interpreter—a set of features that builders can manipulate—quite than a black-box knowledge format managed by the browser.

Strategic evaluation: To undertake or wait?

Pretext is launched underneath the MIT License, guaranteeing it stays a public utility for the developer neighborhood and business enterprises alike. It is not merely a library for making chat bubbles look higher; it is an infrastructure-level instrument that decouples the visible presentation of information from the architectural constraints of the Nineties internet.

By fixing the final and largest bottleneck of textual content measurement, Lou has offered a path for the internet to lastly compete with native platforms when it comes to fluidity and expressiveness. Whether or not it is used for high-end editorial design, 120fps virtualized feeds, or generative AI interfaces, Pretext marks the second when textual content on the internet stopped being a static doc and have become a really programmable medium.

Organizations ought to undertake Pretext instantly in the event that they are constructing “Generative UI” or high-frequency knowledge dashboards, however they need to accomplish that with a transparent understanding of the “thick shopper” trade-off.

• Why undertake: The transfer from O(N) to O(log N) or O(1) format efficiency is not an incremental replace; it is an architectural unlock. In case your product entails a chat interface that stutters throughout lengthy responses or a masonry grid that “jumps” because it calculates heights, Pretext is the resolution. It permits you to construct interfaces that really feel as quick as the underlying fashions are changing into.

• What to pay attention to: Adoption requires a specialised expertise pool. This is not “simply CSS” anymore; it’s typography-aware engineering. Organizations should additionally bear in mind that by shifting format into userland, they turn into the “stewards” of accessibility and customary habits that the browser used to deal with without cost.

In the end, Pretext is the first main step towards an online that feels extra like a sport engine and fewer like a static doc. Organizations that embrace this “interpreter” mannequin of format will likely be the ones that outline the visible language of the AI period.