← back
The Agentic AI Engineer - Benedikt Sanftl, Mutagent
Takeaway
Automating the agent-development feedback loop requires explicit specifications, meaningful evals, and production-grounded diagnosis.
Summary
- Mutagent applies agentic loops to agent development, connecting offline build-and-evaluate cycles with production monitoring and improvement.
- The proposed stages are specification, implementation, evaluation, deployment, monitoring, diagnosis, and targeted optimization.
- Specifications define responsibilities, success criteria, context, integrations, and boundaries independently of the chosen framework or harness.
- Production failures undergo structured root-cause analysis before changes are evaluated and redeployed; existing agents enter through optimization rather than cold-start design.
agent-developmenteval-drivenoptimization
Original description
In this video we introduce the concept of the agentic ai engineer. similar to coding agent loops for agents we build a system that build AI Agents in an Eval-Driven Developement Loop. The Agentic AI Enginner is a collection of a multi-agent team, steared by an orchestrator and combines, spec, build, evaluate, diagnose, monitor, optimse. We round up the talk with a live demo from one of our agents in research preview. Speakers: Benedikt Sanftl (Mutagent): Bene is the CEO and Co-Founder of Mutagent. The platform for Agentic AI Engineering. LinkedIn: / benedikt-sanftl-294a6039a