The A to D in DMTA (Design, Make, Test, and Analyze): Agentic Analysis for Self-Driving Labs

In future self-driving laboratories (SDLs), AI agents will analyze and plan experiments with progressively less intervention, freeing domain experts from repetitive analytical tasks to focus on creative scientific questions. This autonomy must be earned: flexible reasoning must be integrated with validated execution, a traceable record of the experimental state, and expert oversight. We present an architecture that meets these requirements at the Analyze–Design interface of the Design–Make–Test–Analyze (DMTA) cycle, where the agent proposes measurements and interprets results, but every state-changing action passes through authorization and validation. The architecture is implemented in a chemistry SDL and tested in an oxidant screening campaign for the synthesis of an organic dye. The agent selected samples and measurements from liquid chromatography (LC) runs and reasoned over mass-spectrometric (MS) and diode-array detection (DAD) readouts; it identified sample-transfer and measurement-quality failures during physical operation and determined the best-performing condition under finite experimental budgets. Our results demonstrate a deployable design to integrate agentic reasoning in SDLs that is extensible to more complex analytical and synthetic workflows while preserving validation, traceability, and human control.