What Building Slater Search Taught Jared Malan About the Next Layer of the Web
TL;DR
- Infrastructure that makes AI useful is becoming the real advantage in web development, ahead of just building faster with AI.
- Vibe coding works well for internal tools and simple sites, but the process breaks down once more than one person needs to manage the same project.
- AI search rewards sites built around clean data and embeddings (an AI-friendly compression of a page’s meaning). Old SEO tricks like self-serving listicles or paid placements don't carry over.
Most AI-optimized sites are still built backward: a working site first, an AI layer bolted on after. Jared Malan built AnswerCloud, an Edgar Allan product, to test what happens when you build it the other way around, and the answer says a lot about where web development is headed.
Jared and I have been building things together since before either of us knew what vibe coding was. We worked out of a shared space in Atlanta years ago, him running a Ruby on Rails shop, me on the design side. He eventually joined Edgar Allan full time, where he now builds Edgar Allan's developer tools: Slater, Slater Search, and AnswerCloud.
As part of our ongoing Building the Next Web series that’s included conversations with Finsweet’s Jesse Nieman, StackedGTM’s Josh Grant, and Island’s Ari Yablok, I sat down with Jared to talk through what he's built and what he's learned from vibe coding an entire homepage in a week. The bigger question we kept coming back to: is the next fight in web development really about who builds fastest, or about who builds the layer everyone else builds on top of?
The real gap in Webflow has always been JavaScript
Slater, Edgar Allan's Webflow development tool, started as a fix for a specific, miserable workflow. Jared was running a team of no-code developers on Webflow, and any time they needed custom JavaScript, the process was brutal: copy in code, save, publish, wait a minute, check the browser, repeat. For a language that requires constant small adjustments, that loop destroyed productivity.
Jared built Slater to give Webflow a developer-friendly environment: write JavaScript, save, refresh, see the result immediately. It launched at the end of 2022 and picked up real traction through 2023, right as GPT-3.5 was becoming useful enough to write a lot of that JavaScript for people who weren't developers at all. The tool didn't just speed up Jared's own work. It let his no-code team take over tasks that used to land on his desk, freeing him up for harder problems.
Search data is a data problem before it’s a UX problem
Slater Search, Jared and Edgar Allan's newer product, comes from a similar instinct: identify the gap, then build the tool that closes it. Webflow's native search is fairly basic. The alternative, however, something like Algolia, is enterprise-grade and often more than a marketing team needs. The real blocker was always data access, not search functionality. Webflow has always had solid APIs, but using them required more technical depth than most no-code teams had.
Slater Search closes that gap by exposing clean, structured endpoints: raw JSON for developers, a no-code data-attribute layer for teams that want it visual. From there, teams can weight collections and keywords so the most important content surfaces first, layer in custom JavaScript filtering, or turn on AI semantic search, which answers a real sentence instead of matching stacked keywords. That last part matters more than it sounds. People increasingly search the way they'd ask a person a question, not the way they'd type into a 2015 search box.
AnswerCloud, Edgar Allan's answer engine, works as a hosting layer built from the start, so an AI system can find, understand, and cite what's on a site the same way a person would.
An AI-native site isn't one with a chatbot bolted on after launch. It's one where every page already has an embedding (a compressed representation of what the page means, not just the words on it, so AI can match it to a question by meaning instead of keywords), a text version, and a way for an agent to cite it correctly, before a single visitor asks it a question.
Every page on AnswerCloud gets that embedding, an HTML version, and a text version, plus an automatically generated llms.txt file for the agents that come looking. Site owners can see which AI bots are visiting, which pages are answering questions, and where the resulting traffic goes. It's the same shift we talk about constantly in our own Visibility Engineering and Optimization approach: brand, content, and technical structure aren't separate problems. They're one system, and AI search treats them that way whether you've planned for it or not.
Vibe coding is incredible until more than one person needs to touch the site
Jared built the entire homepage for AnswerCloud, Edgar Allan's answer engine, by vibe coding it, largely inside Claude Code, translating a designer's Figma file into a working site in about a week. He's candid about both sides of that experience. When it works, it's remarkable: you can get to an 80% solution almost instantly, and for an internal tool or a project only you touch, that's often enough.
The problems show up the moment a project needs more than one person, or more polish than a single prompt can deliver. Jared's framing: vibe coding is like sending instructions to an offshore developer you've never met. You ask for a change, and you don't always know what else moved. Without real version control, a fix to the homepage can quietly break the pricing page. Tools like GitHub can catch that, but they're built for developers, not for a twelve-person marketing team.
That's why Jared thinks Webflow's move upmarket, toward larger, more complex organizations, makes sense even as AI makes small sites easier to build without it. A local restaurant with twenty pages and an occasional blog update doesn't need Webflow anymore. A company with twelve stakeholders all wanting input on the same site absolutely does, because you cannot have twelve people prompting an AI about the same page without ending up with a Frankensteined mess.
The middle layer is where the disruption lands
The most interesting part of our conversation was about where value is going to sit once AI makes building nearly free. Jared's read: pure infrastructure, the hard stuff around data access, security, and reliability, stays valuable, while simple sites that any team can one-shot with a prompt become commodities. The squeeze lands on everything in between, the traditional SaaS layer that's too complex to vibe code reliably but not differentiated enough to survive once alternatives are a prompt away.
We've both seen the pattern play out on LinkedIn: a team vibe codes its own ticket tracker, celebrates for a month, then quietly migrates back to Linear once they realize they're spending more time managing the tool than the tickets. Ephemeral software works well for internal, single-owner tools, but teams that need to run something together over time still need more structure than that.
On the AEO side, Jared's blunt about what doesn't work either: flooding the web with AI-generated content until Google catches on and tanks your rankings, writing your own "best of" lists and quietly including yourself, or paying to appear on someone else's listicle. All of it works for a while, but none of it holds up, and all of it is exactly the kind of shortcut we'd tell a client to avoid in a Webflow AEO audit.
Know which problem you have before you pick a tool
If you're a single owner solving a contained, internal problem, vibe coding is the fastest and cheapest way there, and Jared would tell you the same thing. If more than one person needs to manage the outcome over time, you need the structure a platform provides, paired with the kind of brand clarity assessment that tells you what the site needs to say before anyone touches a tool.
This kind of thing is definitely on our minds here. It’s the same instinct behind ag-nts, Edgar Allan's measurement layer for AI agent traffic on Webflow, and behind the way we think about our Customer Success Program (CSP): the tools change fast, but the discipline of knowing what you're building and why doesn't.
Products mentioned
- Slater
Edgar Allan's developer environment for writing and testing custom JavaScript inside Webflow, without the publish-and-wait cycle. - Slater Search
Structured, developer-friendly search endpoints for Webflow sites, built by Edgar Allan, with optional AI semantic search built in. - AnswerCloud
Edgar Allan's answer engine hosting layer. Gives every page an embedding, an HTML version, a text version, and an llms.txt file so AI systems can find and cite it. - ag-nts
Edgar Allan's measurement layer for tracking AI agent traffic on Webflow sites.
FAQs
What's the difference between vibe coding a website and building on a platform like Webflow?
Vibe coding means prompting an AI to generate a working site directly, often through a tool like Claude Code, with no visual editor in between. Webflow gives you a shared, structured environment where changes are visible and multiple people can work on the same project without stepping on each other. Vibe coding is faster for a single owner on a simple project, while Webflow holds up better once a team is involved.
Is vibe coding good enough for a real business website?
It depends on who's using it and how many people need to manage it. For a small business with a simple site and one person handling updates, vibe coding can get you most of the way there quickly. For a larger organization with multiple stakeholders and ongoing changes, the lack of version control and shared structure tends to cause more problems than it solves.
What is AEO and how is it different from SEO?
AEO, or answer engine optimization, is the practice of structuring a site so AI systems like ChatGPT or Perplexity can find, understand, and cite its content accurately. SEO optimizes for traditional search rankings, while AEO optimizes for being the source an AI pulls from and names when it answers a question, which depends more on clean data, embeddings, and clear structure than on keywords.
Do I need a separate AI-optimized version of my website?
Not necessarily a separate site, but you do need content structured so AI systems can parse it well: clear data, text versions of pages, and files like llms.txt that tell agents what's available. Edgar Allan built AnswerCloud to handle this from the start rather than bolting an AI layer onto an existing site after the fact.
How do I know if my site is showing up in AI search results?
You need visibility into which AI bots are visiting your site and which pages they're pulling answers from, similar to how Edgar Allan's ag-nts tracks AI agent traffic on Webflow. Without that data, you're guessing whether your content is being cited or simply sitting unseen.
When does it make sense to move off vibe-coded tools onto something more structured?
The signal is usually organizational, not technical. Once more than one person needs to make changes, or once you need confidence that a fix in one place won't break something else, you need real version control and shared structure. That's the point where a platform like Webflow, paired with a clear content and brand strategy, starts to earn its cost.