Science in the Agentic Era: Structured Experimentation With Ado
Outlines science in the Agentic Era: Structured Experimentation With Ado and evaluates key implications for security practitioners.
Summary
Coding agents
Alessandra Pomp defines coding agents as LLMs with harnesses for tool use, state, and iterative explore–plan–execute–verify loops—mirroring DevOps and the scientific method.
Science parallel
Research follows the same cycle: question/explore, hypothesis/plan, experiment/execute, analysis/verify—suggesting agents can support rigorous science, not only software delivery.
ADO introduction
ADO (Accelerated Discovery Orchestrator) extends coding agents for science via Pydantic schemas, provenance tracking, structured result stores, plugins, Ray-backed scalable execution, and a CLI usable by humans and agents.
Workflow
Agents retrieve plans and configurations, define discovery spaces, run experiment campaigns, store/query results over time, and verify outcomes through schemas, logs, and sampling trajectories for troubleshooting.
Related talks
The Linux Foundation·43 min
Eval-Driven Development: Mastering Agentic Tracing and Expert-Aligned Judges With MLf... V. Caldeira
Discusses eval-Driven Development: Mastering Agentic Tracing and Expert-Aligned Judges With MLf... V. Caldeira and evaluates key implications for security practitioners.
The Linux Foundation·31 min
Kevin Dubois · Daniel Oh
Details building and Orchestrating Production-ready Agentic AI… and evaluates key implications for security practitioners.
The Linux Foundation·32 min
Aswath Srinivasan
Explores personalisation and Specialisation of Search With OpenSearch Agentic Search and evaluates key implications for security practitioners.