From 24% to over 80%
SIRPA Terre wanted to move to generative AI for real files, real memos, and real production schedules, in an institution where procedures matter and confidentiality is not an option. In six months, the department went from experimental use to a galaxy of 21 operational AI agents. Today, the teams design their own agents, without us.
Our challenge
- Bringing AI into a structured framework: integrating a tool as fluid as AI into a military communication department without disrupting existing workflows.
- Avoiding the Phase 1 trap: trained points of contact but without leadership support, resulting in adoption capped at 24%.
- Building tools that are actually used: tools that people use on a Tuesday morning, not just on demo day.
Approach
Map before you train
Understand how teams draft speeches, monitor the press, or produce visuals for social media. AI doesn't replace a workflow; it plugs into it.
One agent = one assisted expert = one skill
Ultra-specialized agents that hand off tasks to one another (media monitoring, drafting, proofreading, social media production), spread across five areas: writing, imagery, media monitoring, strategy, and a secure internal section for sensitive topics. Humans retain decision-making authority.
Changing strategy mid-stream
Phase 1 plateaued at 24%: the points of contact were skilled but isolated. We switched to workshops by business unit, with leadership support. Adoption rose to over 80%.
Train to use, then to create
Eight people were trained as AI ambassadors; two became the department's operational AI leads. They are the ones who design, update, and pass on the agents today.
What this case reveals
A successful AI deployment isn't measured by the number of agents installed, but by the number of days the AI runs without needing attention. SIRPA Terre no longer needs Merkutio to keep its ecosystem running: that is exactly the result we aimed for.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Suspendisse varius enim in eros elementum tristique. Duis cursus, mi quis viverra ornare, eros dolor interdum nulla, ut commodo diam libero vitae erat. Aenean faucibus nibh et justo cursus id rutrum lorem imperdiet. Nunc ut sem vitae risus tristique posuere.
Let's talk about your
next challenge.
