Patryk Radtke

Let's talk.

Send opens your email app with this message ready to go.

Home
Projects
About
LinkedIn
© 2026 Patryk Radtke. All rights reserved.TermsPrivacy policy

Zest · 2025 → present

Zest: AI adoption analytics for engineering teams

See live site

I designed the brand, the product interface and the marketing site, and built the site in code.

~5 days

from the first design to a live marketing site

Design start to production deploy

51

screens in one design system, across web, VS Code, Chrome, Slack and the CLI

Distinct app routes in the Zest codebase

Overview

Engineering teams pay for AI coding tools but have no clear view of what gets used, what works and whether the cost pays off. Zest is a privacy-first analytics layer that answers those questions. My job was to make a complex, abstract product easy to read, and then help ship it.

The problem

Engineering leaders had no way to measure AI adoption. Their dashboards showed what developers shipped, but not how they used AI to get there. The concept was also hard to grasp: a five-level AI maturity model that covers individuals, teams and the whole company, presented on a generic SaaS template that explained none of it.

  • Three audiences in one interface: developers, team leads and executives
  • A maturity framework that was hard to picture
  • A lot of usage data that becomes noise when you show all of it at once
  • A marketing site that looked like every other dev-tools template

The solution

I organised the product around a personal AI Toolkit. Developers see their own tools, skills and agents first, team leads get leaderboards and AI standups, and executives get adoption trends and cost governance. All three views share one visual language, from a single developer's week to the whole company's spend. For the marketing site I turned the five-level framework into an interactive product story and built it myself in Astro, React and Tailwind, so the live site matches the design.

  • A per-developer AI Toolkit that shows detail only when you ask for it
  • Leaderboards and auto-generated standups a lead can read before a call
  • Spend, seat and token governance, so executives can see the return on the cost
  • One identity across web, a VS Code side panel, a Chrome command menu and Slack
Product film, with sound
Team view to AI standup
Ask Zest: a playbook run
A developer's own week: AI tasks, hours, PRs opened and merged, plus an auto-generated standup and the weekly "AI meme".
The shared feed: GitHub cycle-time metrics pulled in via MCP, and an agent-written "PR hero of the week".
A verification session expanded inline — its type, findings and the model stack behind a single PR.
The AI stack behind one person's score: platforms, models, skills, CLI and MCP tools, with session usage.
Reusable AI standups, cheatcodes and prompts the whole team can share and version.
A connected agent reads AI spend across coding tools and points to where the team can save.
Monthly cost by platform, seat utilisation and optimisation tips for whoever owns the budget.
Plan limits and active hours per engineer for one coding-agent subscription.
Every scheduled report across the workspace: owner, destination, trigger and last-run status.
One report up close: its goal, schedule, a live content preview and its delivery history.
New members inherit the right reports for their job: coding, analyst, manager, PM or designer.
Building teams from repositories and choosing the columns each leaderboard will track.

Process

Discovery

Stakeholder sessions

Three audience definitions

Maturity model mapping

Design

Design system and tokens

Dashboard and data-viz patterns

Cross-surface UI

Build

Astro + React + Tailwind

Image and LCP pass

35+ preview deploys

Result

Zest now uses one design language across four surfaces, and new features are built inside it. The marketing site went from a generic template to an interactive walkthrough of the maturity model, which I built and deployed without touching the client's production infrastructure.

I still review the live product in PostHog session recordings and turn what I see into UX reports, so each change I propose is backed by recordings of how people use it.

  • One design system now covers 51 screens across web, both extensions, Slack and the CLI
  • Marketing site designed and shipped to production in under a week
  • Ongoing UX reports built from live session recordings
Next project: ePlanAll projectsHome