Judge the Judge: Building LLM Evaluators That Actually Work with GEPA — Mahmoud Mabrouk, Agenta AI
AI Engineer Europe 2026 · 40:51
Open-source agent workspace
Agenta builds an open-source agent workspace for teams to create AI agents through chat, share them, and automate recurring work. Agenta 2.0, introduced in 2026, extends its earlier prompt management, evaluation, and observability platform. Users can connect applications, maintain documents and research in persistent workspaces, and run agents on schedules or application events. Available through Agenta Cloud or self-hosting, it serves work across engineering, customer support, sales, and operations. Tracing, configuration versioning, and human approval controls help teams inspect failures, refine behavior, and decide which actions require oversight.
Founded in Antler’s Berlin residency in 2023 by Akrem Abayed and Mahmoud Mabrouk, Agenta is led by Mabrouk as CEO. Its engineering approach combines shared workspaces with portable agent configurations built from AGENTS.md, skills, tools, and MCP servers. Teams can reuse compatible components and switch supported models or execution harnesses without rebuilding their agents. The same agent can collaborate with a person in chat and handle recurring tasks in the background.
Antler confirmed its investment in Agenta in 2024 without disclosing an amount. By August 2026, the project’s GitHub repository displayed approximately 4,600 stars, a measure of community interest. Agenta offers both a hosted service and an MIT-licensed self-hosted product, allowing organizations to run agents and retain data on their own infrastructure.
AI Engineer Europe 2026 · 40:51
Affiliations reflect their AIE appearances, not necessarily current employment.
The workshop connects evaluator validation with a production data flywheel, using observability to inform continuous evaluation improvements.
Affiliations reflect each recorded session, not necessarily current employment.