Summary
Prosus described how it encouraged 40,000 employees to build agents, yielding about 60,000 agents, 24,000 used continuously, and roughly 500 that “pay the bill” for the rest. The panel emphasized measuring ROI, doubling down on high-value agents, and using a portfolio of custom, distilled models with Nebius infrastructure to cut agent interaction costs by 90% ahead of projected 20x token growth.
Key Topics
5 key topics from Danila Shtan, Euro Beinat and Harry Booth at HumanX Amsterdam 2026. Thicker branches were mentioned more often in the talk.
Beyond the Pilot: Making AI Agents Work at Scale
Click a topic to open it
Key Takeaways
- Encourage employees to build agents, then identify workflows with measurable ROI.
- Measure agent impact across efficiency, quality, agility, and cost before scaling.
- Use model portfolios instead of premium models for every agentic task.
- Distill domain models into smaller production models for lower-cost interactions.
- Feed user signals back into models to reduce drift and improve relevance.
“we need to make sure that or encourage that everybody in our organization, about 40,000 people, everybody becomes a builder.”
Share
Up Next
NH
WatchNext in agenda
Europe, Sovereignty & the AI Race
Nick Clegg, Hadas Gold
ai regulationeuropean ai
GC
WatchAlso on ai agents
Leapfrogging from Chatbots to Autonomous Workflows
Gautier Cloix, Charlie Perreau
ai agentsautonomous workflows
DJL
WatchAlso on ai agents
Rewriting the Operating Model for the AI Era
Des Traynor, Jennifer Smith, Luisa Bomke
ai agentsenterprise ai