AI Council — Design & Development

From Figma to Terminal

AI Council (formerly Data Council) is a technical conference with over 13 years of history, bringing together the engineers who build the systems the world runs on, now focused on the AI stack. Litebox partnered with the AI Council team on brand direction, the website, key visuals, and a wide range of animations for web and social.

This case study isn't really about the whole site. It's about the point in the project where I stopped handing off designs and started building them myself.

Starting the Traditional Way

Like most projects, this one began in Figma: structure, layout ideas, the site's overall direction. Nothing unusual yet, just the groundwork every site needs before anyone touches a single line of code.

Where It Got Interesting

Once the direction was set, we switched into Litebox's Direct Design workflow: designing straight into code with Claude and Claude Code, using the Figma file and the project's key visuals as the single source of truth. No handoff, no waiting on a dev to interpret a static file. Faster iteration, real time visual and motion testing, and a much tighter match between what we designed and what actually shipped.

It also meant learning GitHub from zero, in the middle of a live project: setting up an account, understanding branches, commits, push, pull, figuring out how to actually version and publish a site instead of just exporting a file.

What I Actually Built

Working hands on with Claude Code, I was directly responsible for several pieces of the final site, all implemented and animated in code, mostly CSS:

The loading screen.

The interactive footer icons.

And several of the site's transitions and animations throughout.

What This Project Actually Taught Me

This is the project where design stopped ending at Figma for me. Learning to move between a design tool, a terminal, and an AI agent to build something real, not just illustrate it, changed how I think about my own role. I went from handing off a file and hoping it survived translation, to being the one making the final call on how something actually behaves in the browser.

It's also where I started paying real attention to AI as part of my process, not as a gimmick, but as a working method: how to prompt an agent well, how to iterate with it in real time, how to trust it with real implementation instead of just using it for inspiration or copy. Git, branches, and Claude Code went from things I'd only heard about to tools I now reach for by default.