--- title: Medium source: Wix Engineering source_slug: wix url: https://medium.com/wix-engineering/how-a-tool-i-built-for-my-team-became-wixs-official-bi-platform-ca206e479262?source=rss----e239c562f907---4 published: 2026-07-28 fetched: 2026-08-04T14:35:47+00:00 ingested: true --- # How a Tool I Built for My Team Became Wix’s Official BI Platform Press enter or click to view image in full size This past year, AI shifted what we expect from our tools. Once you’ve seen an agent build a complex feature for you in seconds, going back to a tool that forces you to do everything by hand feels almost unbearable. Nobody wants to use software without AI anymore, and that goes double for dashboards. Because suddenly, “building” a dashboard got easy. And the moment it got easy, everyone wanted more of them, faster and they wanted to build their own. **Vizion** started as my attempt to keep up with that demand without drowning in manual work. I tried the obvious thing first: I asked an AI to build me a dashboard. It did it instantly, and it was beautiful. It was also completely empty, a perfect picture of numbers that didn’t actually exist. That gap between something that **looks** real and something that **is** really turned out to be the whole story. **The challenge was never the HTML, it was trust.** Generating the interface is the easy part now. An AI will hand you ten beautiful dashboards before lunch, the hard part is the question underneath: can you trust the number on the screen enough to make a decision with it? Because an AI will give you a confident, beautiful, wrong number and never warn you. Nothing crashes, there’s no error on screen. The outdated chart looks exactly as trustworthy as the correct one next to it. And a dashboard that’s quietly wrong is worse than no dashboard at all, because someone will believe it. **The Missing Link** People were already generating dashboards on their own. They’d get something that looked great and then hit a wall, because it lived on their laptop and was wired to nothing. The frontend was never the gap. The gap was everything between that local mockup and a real dashboard running on live company data. In Wix we already had every hard piece to close it. **Trino** could query every data source we own. **MCP** could give an agent the context around the data and not just the data itself, so it understood what a table meant instead of guessing. **Serverless** could turn all of that into a running service in an afternoon. None of it had ever been pointed at the same problem. So I built a proof of concept on top of all three. You describe what you want in plain language, an agent behind multiple skills writes the SQL grounded by the context MCP hands it, the query runs on live data through Trino, and the result comes back as a real dashboard. It came together over a few evenings, and it worked far better than it had any right to. I never announced it. I showed it to a handful of people, they built something useful, others saw it over their shoulder, and within weeks it was spreading on its own. Today it runs across more than **50** groups, with over **2,000** views a day and **1,000** monthly active users, and I had almost nothing to do with that beyond getting out of the way. Press enter or click to view image in full size **Then everyone started asking for more** I put a small button in the product that let anyone request a feature, and they used it constantly. What surprised me was how the loop felt from my side. It was almost always faster to ship the feature than to write back and explain when we would get to it. Someone asked in the morning, and the honest reply was the thing itself, live, by the afternoon. Whether it is scheduling, so a dashboard was refreshed and waiting before you opened it. Email/Slack, so it came to you as a screenshot. Export to CSV, Search, filters, and a long tail of “could it also do this” that never stopped. Building had become so cheap that requests stopped being a backlog and turned into a conversation. That was the moment Vizion stopped feeling like a tool and started feeling like a place to create. **Teaching the AI to build like us** Letting an AI write raw HTML and CSS gets you something different every time, costs tons of tokens, and most of it is fragile. A chart that works in one dashboard breaks in the next. Nothing looks like it belongs to the same product. So we did the opposite of freestyling. **We designed a component library** , around thirty tested, branded building blocks made specifically for the AI to choose from: KPI cards, data tables, filter bars, date pickers, sparklines, and a dozen chart types from line and bar to funnel. Each one is already styled in Wix language and already handles its own loading, error, and empty states. Now the agent doesn’t invent a chart, it picks one. Building a dashboard became assembling known-good pieces instead of generating fragile markup, which is exactly what an AI is reliable at. Every dashboard came out consistent and on-brand, the AI’s job got easier, and a whole category of “looks fine, actually broken” simply stopped happening. Press enter or click to view image in full size **Built to be trusted** When anyone can generate a chart, the chart stops being the achievement. The only thing left worth engineering is whether a person can trust it. So I stopped writing cleverer skills and built the trust **into the pipeline** , in the places the AI couldn’t skip. Every dashboard eventually is just files in Git: a manifest that describes it, the SQL behind it, and the page itself. Every change, whether a person or an agent made it, runs through the same automated checks before it can go live. It’s a code review that never gets tired. The manifest has to declare what it should, every SQL placeholder has to be accounted for so a query can’t run with a missing filter, components should come from the one shared library, and most important of all, every query has to actually execute against live data and return something sane. Nothing ships without passing CI first. Press enter or click to view image in full size Underneath each team is its own independently deployed service, running on its own Kubernetes pods, that the central hub discovers and stitches into one experience. The isolation is the point: a team can ship, break, and fix their own dashboards on their own schedule without ever touching anyone else’s. When you publish, you open a pull request, the checks run, and a real problem comes back as a violation with a plain explanation of how to fix it. If everything passes and it’s just a dashboard, it approves and merges itself and deploys automatically, live a few minutes later. Changes to the platform itself still wait for a human, and anything touching sensitive data has to name who can see it or it fails closed. The quiet trick in all of it is that someone who just wanted a funnel chart gets the full discipline of professional software delivery, version control, continuous integration, isolated deployments, automatic review, without ever learning that any of it is there. And when it gets messy we have skills for “Creating a PR” and “Fix PR comments”. **How it works?** We didn’t invent a new way to code; we just built a bridge. We took all that heavy, invisible complexity the Git workflows, the automated tests, the production deployments and wrapped it in plain, conversational English. Instead of making the user worry about where the data lives or how the code is structured, the system handles the heavy lifting, ensuring that every request triggers a rigorous, professional pipeline before anything ever hits the screen. It turns out that when you remove the friction of the ‘how,’ the entire experience changes. You aren’t just typing into a black box anymore; you’re triggering a pre-configured build process that guarantees you won’t accidentally break anything. By pulling the curtain back on the complexity, you’re left with a system that feels less like a tool you have to manage and more like an extension of your own thoughts. **From one laptop to Wix’s BI platform** At some point a side project spreading on its own becomes too important to stay a side project. So we assembled a real team around it, and for one focused month a group of engineers, designers, and data people worked on almost nothing else, hardening the platform and pushing the performance until it could carry the whole company without flinching. Then the thing I never let myself imagine actually happened. Vizion became Wix’s official BI platform and replaced the main BI tool we’d leaned on for years. That old tool ran on licenses that cost the company around half a million dollars a year, one to build a dashboard and often another just to look at one. All of it is gone now. A side project quietly erased a half-million-dollar line item, and nobody ever asked it to. But the money was always the easy part to measure. The real change is harder to put on a slide. **Handing the craft back to you** A dashboard used to be a static report something that took days to build, a team to maintain, and was often stale the day it shipped. Now, it’s a living product that evolves the moment you ask it to. But the real shift isn’t just in the technology, it’s in who gets to use it. AI didn’t take away the interesting part of the work, it took the busywork that kept us from it, leaving the part that actually matters: deciding what’s worth showing and ensuring the story it tells is true. This means that whether you are an analyst, a PM, or anyone with a question about our data, you are now a creator. You don’t need to be a data engineer to build professional-grade data products. You just need to have a Vizion. This shift isn’t just about dashboards; it’s about realizing that the gap between a side project and a system that changes how we work is smaller than we used to think. If you have an idea for a system that can solve a real problem and move the needle, you don’t need to wait for a roadmap, you can build it yourself. The tools are already open on your laptop. The hardest part isn’t the architecture or the engineering anymore; it’s just deciding to start. **Huge thanks to the Dev Platform data team, the data engine team, BA guild and everyone who helped push this from a laptop experiment into a platform that carries the company. I can’t wait to see what we build next.**