0

Your cart

TOTAL excl.
TOTAL incl.
Pay
Go back

Can we confidently entrust the updating of 200 heterogeneous and undocumented servers to AI?

REXAutomatisationInfrastructure

2026-12-09 | 11:35 AM - 11:55 AM

Information

Performing updates on 200 legacy, heterogeneous VMs without documentation: mission impossible?

A story of a successful operation thanks to AI, starting from an initial PoC that was too token-intensive, through a controlled hybrid workflow, and then scaling up.

Upgrading machines is rarely a walk in the park. But when each client arrives with their own legacy system, when no two machines are alike, when there is no pre-production environment and — above all — when documentation is scarce, it borders on a nightmare.

In 2026, the idea is appealing: to plug in AI to escape this hell. Except that in practice, the trap quickly snaps shut with overloaded contexts, slow execution, and astronomical token bills.

At Theodo Cloud, we met the challenge by building an agentic factory capable of handling the heterogeneous legacy systems of different clients across more than 200 machines. By moving from a simple proof of concept to a full-scale execution factory, we increased reliability while keeping costs under control.

During this feedback session, we will retrace the complete trajectory of the project:
  • From monolithic PoC to multi-agent: how our v0 (single agent) and our v1 (fleet of agents) validated the concept, but blew our budget.
  • Re-pragmatization through code: the transition to v2 through the reintroduction of deterministic scripts for exploration and testing, coupled with a knowledge base to avoid reanalyzing machines in a loop.
  • Scalability and security: the culmination with v3, its near-autonomous execution on non-production and the maintenance of human validation (*human-in-the-loop*) on production.
At the end of this talk, you will leave with:
  1. A concrete method to reintroduce determinism into your AI agents and to know what to entrust to deterministic scripts rather than to LLM reasoning.
  2. Feedback from our clients on the ground : perception of change, level of confidence and real impact on their teams.
  3. Our actual production metrics : time savings per server, financial ROI and exact percentage of tokens saved from v1 to v3.

Speakers