Leonie Monigatti

Leonie Monigatti is a senior developer advocate at Elastic, machine learning engineer, and technical writer specializing in agentic search and context engineering. She designs practical ways for AI agents to retrieve trustworthy information from databases, files, memory, and the web without wasting context or relying on inappropriate tools.

Trained in electrical engineering, Monigatti previously developed software in the automotive industry. After leaving her corporate engineering role in 2023, she wrote full time before joining Weaviate as a developer advocate. There, her work expanded from educating developers about vector databases and retrieval evaluation into machine-learning engineering and research. She coauthored analyses of late-interaction retrieval models and database querying through function calling. After approximately two and a half years at Weaviate, she moved to Elastic.

• Search tools matched to the task. Monigatti argues that fixed retrieval pipelines fetch irrelevant material and miss questions requiring multiple searches. Effective agents choose among exact matching, semantic retrieval, structured queries, aggregation, and web or filesystem search according to the actual problem. Her agentic-search guide identifies three common failures: agents skip retrieval, choose the wrong tool, or supply incorrect parameters.

• Low-floor, high-ceiling tool design. Specialized tools offer reliable, simple operations; general-purpose database and shell tools handle unfamiliar or analytical requests. Monigatti recommends starting with broader capabilities when usage patterns remain unclear, logging failures, and adding specialized interfaces where repeated behavior warrants them. Shell access requires sandboxing and careful permissions.

• Progressively loaded agent skills. At AI Engineer Europe, she demonstrated how semantic search missed an exact technical term and an agent-generated Elasticsearch query failed after confusing SQL and Elasticsearch wildcard syntax. Loading targeted query guidance only when needed, describing tool dependencies explicitly, and returning recoverable errors enabled the agent to correct its search without stuffing documentation into every prompt.

• ElasticsearchFs. Her open-source proof of concept gives agents filesystem-style access to Elasticsearch data through familiar commands such as ls, cat, find, and grep, combining a shell-like interface with database-backed storage and read-only safeguards.

• Agent Journey Map. Monigatti’s agent-experience framework maps how autonomous systems discover, evaluate, adopt, and use software. It treats documentation, authentication, permissions, command-line interfaces, reusable skills, and human approval as product-design considerations for software increasingly operated by agents.