Autonomous Agents
Agents that monitor, prepare, route, and act without waiting for a prompt. Scheduled analysis, tool use, escalation, and communication.
This is what I build.
Olivier Debeuf De Rijcker
Belgium
I came up through IT, then built an esports talent agency from scratch. That work trained the habits I still use now: move quickly, work from evidence, and stay close to the real decision instead of the abstract plan.
I now work as Chief AI Engineer inside a pan-European consulting enterprise of roughly 2,000 people. I build applied AI systems, agent workflows, and internal tooling that have to work in production. Alongside that, I contribute upstream to the tools I use and still represent select professional tennis players through Surge.
CURRENTChief AI Engineer at a 2,000 FTE consulting enterprise
ARCIT → Esports agency → AI systems
FOCUSApplied AI systems, agent workflows, internal tooling
SURGERepresent select professional tennis players
CONTACTack@olivierddr.com
What I Build
Applied AI in production.
The recurring work sits in three layers: autonomous agents, workflow systems, and internal tooling for the teams using them.
Workflow Systems
AI embedded in the actual process: intake, matching, document review, and multi-step coordination. The workflow matters as much as the model.
Internal Tooling
Interfaces and operator software for people doing the work every day. Clear, fast systems for teams who cannot afford friction.
Proof
Current role, public work, and operating context.
Pan-European Consulting Enterprise
Current role. Chief AI Engineer inside a 2,000 FTE consulting enterprise. Built and shipped an AI-native replacement for legacy software into production.
Seld
Open-source personal AI platform and conversation-first agent console. Multi-channel messaging, persistent identity, and autonomous execution. The trace above runs on it.
Claude Code
Contributed a fix to Anthropic's Claude Code after a bad release left the tool unusable. Merged upstream.
PR #16683 ↗Databricks MCP Server
Built and open-sourced an MCP server for Databricks so coding agents can inspect and operate Databricks environments.
github.com ↗Latent Space Podcast
Podcast appearance on MCP servers, agent architecture, and personal AI infrastructure.
latent.space ↗Surge BV
Through Surge BV, I still represent select professional tennis players. The operating discipline started in esports, where more than $50M was negotiated across contracts before the work extended into tennis.
surge.tennis ↗Approach
Operating principles.
01
Start with what is failing.
Not with what is possible. The interesting problems are always in the gap between the tool and the work.
02
Build the system, not the demo.
Demos are easy. Systems that survive Monday morning are hard. I build for Monday morning.
03
Models do judgment. Rules do structure.
Don't make the model decide what it shouldn't. Don't make rules handle what they can't. Know the boundary.
04
Ship narrow, then widen.
A working system on one workflow beats a prototype across ten. Prove it works, then expand.
05
Write it down so I can leave.
If the system needs me to run, I failed. Documentation, handover, and operational clarity are part of the build.
If there is a real bottleneck, send it.
The best starting point is concrete: what is failing, what is slow, or where the team is losing time.