Experimenting and preparing the scalingA story behind the organisational transformation through 20 AI agents at Luminus.
Problem
Luminus wanted to catch the GenAI wave — not as spectators, but as first movers.
The challenge: turn curiosity into value while staying agile in a field that changes weekly.
Teams were enthusiastic but scattered. Tools evolved faster than roadmaps could keep up.
The mission: create structured experimentation and raise AI maturity through hands-on building.
Vision
Each team to create two AI agents as collaborators:
- One inward-facing (supporting internal team workflows)
- One outward-facing (helping other teams access knowledge and services)
This design reframed AI from concept to colleague — something to build with, not merely discuss.
Problem Analysis
To build AI maturity, people first needed a safe playground. Luminus launched an internal contest, gathering 38 creative ideas ranging from micro-prompts to fully automated flows.
36 prototypes were developed; two were excluded for complexity or customer-facing risk.
🔥 The Experimentation Phase
Prototyping :
On the 36 ideas, 20 proposed ideas have been started to be prototyped. Agents covering the full agentic spectrum — from prompt-driven helpers to complex automation chains with connectors and actions.
My Role:
As GenAI Adoption Engineer, I coached teams, co-designed prototypes, aligned stakeholders, facilitated hackathons, and built the Agentic Design System — a reusable component library for future agents, with guidance, documentation, resources...
Beyond Prototyping
Tools were only part of the equation.
Together, we established the Luminus AI Blueprint
— documentation, governance, and design guidelines defining what a “good agent” looks like.
It provided consistency, credibility, and scalability for future deployments.
💡 The Hackathon Moment
Ten standout agents advanced to a company-wide hackathon.
Ideas became refined prototypes, evaluated and classified as:
- Stop — not viable
- Keep as Copilot — local use
- Go to AI Portal — enterprise-wide deployment
Setting up scaling and industrialisation🔥 Fast and Flexible
AI evolves at hyper speed.
The design system and modular components allowed seamless updates
— teams could swap connectors, update prompts, and integrate new APIs without re-building from scratch.
Outcomes
Cultural Impact
Employees moved from passive curiosity to proactive innovation.
Teams now design and iterate on agents autonomously, with coaching demand tripled since launch.
Results
As the initiative is closing at the moment of redaction, not many quantified results are available, but
- For customer care ambassadors, that use the email channel, it is emails written up to 3 to 5 time faster, for several dozens of ambassadors, dozens time a day.
- For the impact analysis tool:
- we go from 1 week to get the document to 10 minutes. Quality is rising too.
- One of the users said ”I have 20 years of experience doing impact analysis, but the bot clearly does a better job than I do”
💡 Organisational Learning
The initiative produced more than prototypes — it built capabilities.
The AI Blueprint now serves as the foundation for training, governance, and enterprise scaling.
What started as a hackathon evolved into a self-sustaining system of innovation.
🔥 Finally
The initiative produced more than prototypes — it built capabilities.
The AI Blueprint now serves as the foundation for training, governance, and enterprise scaling.
What started as a hackathon evolved into a self-sustaining system of innovation.
💡 Key Takeaways
- Structured experimentation accelerates learning.
- Embedding AI into daily work fosters ownership and engagement.
- A design system for agents ensures scalability and coherence.
- Continuous adaptation keeps innovation friction-free.