The $750K-Trained 'Website Devourer' That Reverse-Engineers Any Site's Design System in 75 Seconds

Source: Y Combinator | Published: 2026-06-19T14:00:26Z

Webflow co-founder Bryant Chou's new product Ploy can deconstruct any website's design system and rebuild all its components in 75 seconds — 12% of the current YC batch is already using it.

Rebuilding a Website in 75 Seconds — But the Website Isn't the Point

Bryant Chou's Light Cone demo opened with a nostalgia segment: he fed the hosts' decade-old startup websites into Ploy and had it redesign them. Garry's 2008 Posterous, Jared's 2007 Scribd, Harj's 2007 Auctomatic — all rebuilt as modern 2026 versions.

Ploy's core technology is called the "Slurper" — the team spent roughly $750K in tokens building this purely deterministic website-parsing system. It doesn't just screenshot and replicate. It deconstructs the target site's complete design system: button styles, type hierarchy, color palette, component structure — extracting everything. All subsequent generation strictly follows this design language, avoiding the "style drift" common with other vibe-coding tools.

But Bryant kept emphasizing: the website is just the entry point. Ploy's real ambition is to become a company's "marketing brain" — plugging into 50-plus tools, connecting GitHub, Figma, CRM, Google Analytics, and Google Search Console, automatically analyzing traffic data every night, checking search rankings, tracking visitor behavior, then proactively surfacing optimization recommendations.

Not Prettier — Finally Comprehensible

The most interesting reactions in the demo came from the hosts themselves. Seeing Ploy's redesign of Auctomatic, Harj said: "I finally understand what Auctomatic did." Diana had a similar reaction after seeing the new version of her 2017 AR company's site.

Garry pinpointed the key difference: it wasn't just a visual upgrade — the content itself got better. Ploy first understands your business context — who your customers are, what problem you solve, what your core value proposition is — then uses that understanding to restructure your information architecture. The most common feedback from teams using Ploy in the current YC batch: "I can finally tell my own story clearly."

This echoes a classic ritual from YC's early office hours — partners would go through each founder's website, helping them figure out how to explain what they do to the outside world. That job has now been handed off to an AI product.

3,500 Curated Prompts to Fight AI Aesthetics

AI-generated websites have a signature look: rounded cards, left-aligned layouts, cookie-cutter gradients — the industry calls them "AI tells." Ploy's countermeasure is an internal design corpus called the "Lookbook," a curated collection of cutting-edge web design backed by 3,500 design prompts.

Bryant used an analogy: good human designers don't create from nothing — they absorb vast quantities of great work, then produce something original. Ploy tries to replicate this process — instead of letting the model generate freely, it uses carefully curated design references to "override" the model's built-in aesthetic defaults.

"In the AI era, you need enough domain expertise to know how to harness the infinite intelligence inside the model. People who've spent a decade or more in an industry — they know how to leverage the model's underlying capabilities to create something truly world-class."

From Webflow's Single Persona to Ploy's "Boil the Ocean"

When Webflow entered YC in 2013, the website-builder space was absurdly crowded — Bryant recalls at least eight direct competitors. Webflow's strategy was extreme focus: serving roughly 50,000 freelance web designers. Co-founders Sergie and Vlad's obsession with pixel-level detail gave the product a "professional feel" that set it apart.

Ploy takes the opposite approach. Bryant calls it a "boil the ocean" product — targeting tens of millions of small business owners and founders, with AI making that breadth possible. In the Webflow era, serving such a diverse customer base was technically infeasible. Now, models' general capabilities let a single product simultaneously understand the different needs of restaurants, SaaS companies, and e-commerce brands.

PG Couldn't Figure Out Webflow Because He Only Knew HTML

Bryant shared a YC anecdote: during 2013 office hours, Paul Graham couldn't use Webflow. PG understood HTML tags and could add font attributes — Hacker News was probably the upper limit of his CSS abilities. The team had assumed someone as technical as PG would pick it up instantly. Everyone broke into a cold sweat.

The story isn't just a fun aside — it led to a key product decision for Ploy. The team initially invested heavily in bringing Webflow-style visual editing into Ploy: panels, dragging, resizing, layout-flow controls. They kept postponing the feature, and eventually realized they didn't need it at all. Give the model enough context — screenshots, images, annotations — and users can just click on a page element and say "make this copy bold."

When Your Competitor Uses Ploy and You Don't

Garry raised a competitive-dynamics point: if Ploy is good enough, companies that don't use it will get crushed on marketing efficiency by competitors that do. It's similar to programming tools today — in 2026, not using Claude Code or Cursor to write code is almost unthinkable.

Bryant's response was revealing. He said software engineers are the most fickle customers — they jump ship the moment a new tool drops, and competition eventually devolves into "who gives more tokens." He'd rather serve small business owners and founders — people with real, ongoing pain points who won't switch tools with every new trend. That conviction came from his early career at Intuit.

Letting AI Agents Register and Use Ploy on Their Own

When asked if they'd considered letting agents use Ploy directly, Bryant said they're working on it — but chose CLI over MCP as the interface. The reasoning: Ploy can do too many things. MCP is better suited for well-constrained scenarios; CLI gives agents more operational freedom.

Garry agreed, noting that CLI is becoming the standard interaction layer for agents — one reason Claude Code outperforms Cursor. The command line gives agents more latitude. If agents can autonomously choose which product to use for a task, being chosen by the agent is itself a massive competitive advantage. Multiple companies in the current YC batch are already building products around this logic.

The Rippling Playbook: Start With the Offer Letter

Garry compared Ploy to early Rippling. When Parker Conrad started Rippling, the first product was an offer-letter generator — sounds tiny, but the offer letter is a new employee's very first touchpoint with a company. From there, it naturally expanded into HR systems, identity management, and eventually became an enterprise operating system.

Ploy follows similar logic: a company's first touchpoint with the outside world is its homepage. From the homepage, it naturally extends into SEO, content marketing, visitor tracking, and CRM. Bryant said they can already run nightly scans of traffic, check search rankings, identify visiting companies, then proactively tell you "someone from Target is looking at your website" or "someone clicked your CTA button."

The Compound Interest of Experience: The Era of 40-Year-Old Founders

Garry used a Dungeons & Dragons analogy: every founder is like a character with different stat allocations. In the past, founders who maxed out technical stats but had zero in marketing struggled to win markets no matter how good their product was. AI tools are filling in those gaps — a 200-IQ non-verbal genius can now tell their story clearly through Ploy.

Bryant used his own experience as a case in point. He said he'd always operated under scarcity — scarce time, scarce energy, scarce capabilities. Now AI lets him clone himself, not just technically but across every aspect of running a company: every call auto-transcribed into CRM, proposals auto-drafted, follow-up emails auto-scheduled. Garry ran even more aggressive math: based on his own AI-assisted coding output, it's the equivalent of having 400 to 1,000 clones of himself working simultaneously.

"You don't have to be 40, but you need taste."

There Will Be More Small Businesses, Not Fewer

Bryant made a prediction: the era of big companies dominating markets may be ending. Small businesses and entrepreneurship will matter more than ever. If that's right, then every small business needs to be discoverable, needs to tell its story well, needs to present its brand professionally — and that's exactly what Ploy is built to solve.

He admitted he can't predict how far models will evolve, but he's convinced of one thing: no matter how general the underlying models become, businesses will always need opinionated, scenario-specific solutions. The gap between general capability and specialized product — that's where Ploy lives.