The skill matrix at itenium is currently an Excel file.
For an IT consultancy, this is obviously unacceptable. The cobbler’s children have no shoes, the mechanic drives a rust bucket, and the IT consultancy tracks skills in a spreadsheet.
Enter SkillForge: our custom app where we “forge our skills.”
Empower your team with continuous learning. Track progress, manage courses, and build skills together.
– CEO Steven Robijns, moments before authorizing $1200 in AI token spend
Six teams. Six ridiculous names. Six different interpretations of the backlog.
| Team | Nickname | Vibe |
|---|---|---|
| Obsidian | The Hallucination Hunters | Chasing phantom features |
| Royal Purple | The Tab Tab Tab Engineers | Automation over hesitation |
| Teal | The Prompt & Pray Engineers | Minimal tokens, maximum hope |
| Crimson | The Yolo Deployers | 27,000 lines of code. 18 features. |
| Emerald | The Token Burners | Burned bright. Won everything. |
| Midnight Blue | The Main Branch Pushers | No Staging. No fear. |
Everyone participated in this vibe coding exercise. And by everyone, we do mean everyone. The backoffice staff. The people who normally recoil at the sight of a terminal. The ones who think “git” is something you say when frustrated.
What could possibly go wrong.
They all got Claude Code accounts. They all started prompting. Some of them even merged code.
The developers watched in a mixture of horror and fascination as non-technical colleagues casually asked Claude to “add a button that does the thing” and… it worked?
Sort of.
We set up a React 19 + .NET 10 + Postgres application. The stack was modern, the ambitions were high, and the Excel file was about to become obsolete.
SkillForge was designed to be the ultimate L&D platform for consultants, the PRD had 43 functional requirements:
The whole thing was vibe coded with Claude Code, but we’re not complete animals. As responsible adults, guardrails were set up to prevent the AI from shipping absolute chaos to production.
eslint/eslint: Find and fix problems in your JavaScript code.
typicode/husky: Git hooks made easy. Pre-commit validation.
microsoft/playwright: E2E testing. Because someone has to check if this works.
prettier/prettier: Prettier is an opinionated code formatter.
webpro-nl/knip: ✂️ Knip it before you ship it!
vitest-dev/vitest: Next generation testing framework powered by Vite.
nunit/nunit: NUnit Framework
nsubstitute/NSubstitute: A friendly substitute for .NET mocking libraries.
testcontainers/testcontainers-dotnet: A library to support tests with throwaway instances of Docker containers
dotnet/roslynator: Roslynator is a set of code analysis tools for C#, powered by Roslyn.
The idea was simple: Claude Code does the heavy lifting with a very small CLAUDE.md instructing it to implement everything in red/green TDD. Every PR must pass linting, tests, and code review. In theory, this prevents disaster.
In practice… well, …
Because what’s a competition without a scoreboard… Teams could watch their token count tick up in real-time. Nothing motivates quite like seeing your AI bill climb while your feature count stays flat.
We vibe coded the dashboard, with the Competence Coaches improving the dashboard during the bootcamp, which resulted in some weird discrepancies like the biggest commit (lines added) being bigger than the total amount of lines added of the entire team, and also the dashboard being down for several hours. Maybe we should have had an AI harness for the dashboard as well? Naah…
Total burned: ~$1200
That’s roughly $35 per person. Or, as we prefer to think of it: the cost of one really awkward team dinner, but instead we got a functioning application and excellent blog post material.
As can be seen on the dashboard, Onur was leading in the “most tokens burned” category, he was one of the participants whom took the “Tips & Tricks” from the intro powerpoint to heart and was using git worktrees.
The rest of his team had other ideas on what to do while waiting for Claude Code to finish hallucinating non-existent APIs however, they created a few mini games!
A frantic game where you try to ship features while everything is on fire. Surprisingly accurate simulation.
You can play Deploy Panic @ itenium right now!
An endless runner where you dodge obstacles. The obstacles are bugs. The runner is your sanity. Play Itenium Runner now!
Others had even other ideas on what to have Claude do, including adding memes to their SkillForge application. I didn’t even know Claude could create memes ;)
Team Emerald, The Token Burners were selected by Claude as the winning team based on evaluate-teams.md which was, obviously, also written by Claude. The snake eats its own tail.
Emerald is the clear winner with near-complete PRD implementation. They nailed the coaching workflow: goal assignment with deadlines, readiness flags that track aging (days since raised), live session mode with focused validation view, and immutable audit trails. The CSV-based skill seeding is elegant and production-ready. Coach dashboard includes 21-day inactivity detection with visual highlighting. Activity history tracks validations, goal creation, flag raises, and resource completions as a timeline.
They went home with the Certificate of Legendary Achievement
The final dashboard was very Crimson colored with most commits, largest commit and most churn, somehow they still managed to score the least amount of points on the team evaluations. We’re not sure how it happened, but it’s just not possible that Claude made a mistake somewhere so, yeah, …
| Place | Team | Score | Verdict |
|---|---|---|---|
| 🥇 | Emerald | 169/200 | Winner – Only feature-complete implementation |
| 🥈 | Obsidian | 118/200 | Strong foundation, needs goals/coaching |
| 🥉 | Teal | 114/200 | Best code quality, needs more features |
| 4th | RoyalPurple | 112/200 | Great skill data, stubbed coaching |
| 5th | MidnightBlue | 107/200 | Clean skills/resources, no goals |
| 6th | Crimson | 99/200 | LMS foundation, not a coaching platform |
Can you vibe code a production application in one day?
It Depends :)
With the right guardrails (linting, tests, PR reviews), AI assistance (Claude Code), and motivated teams (competition + leaderboards), you can go from “Excel skill matrix” to “functional React/.NET application” in about 8 hours.
But was it worth the tokens?
We got an app, a blog post, the memory of watching non-technical staff successfully merge code, new memes and 2 mini games.
So yes, definitely ;)
The parts Claude couldn’t generate:
1 / 6Click to view all 6 photos
Other interesting reads
itenium-be/Bootcamp-AI: Itenium bootcamp 2026 – L&D Platform creation source code
itenium-be/Bootcamp-AI-Dashboard: Dashboard for the 2026 itenium AI SkillForge Bootcamp: Hall of fame, team statistics
itenium-be/Bootcamp-AI-Resources: Team pictures, certificates, intro powerpoint, …
