A unified AI-agent-assisted platform for coupled experimental and simulation-based workflows in self-driving laboratories

Self-driving laboratories (SDLs) promise to accelerate experimental research, yet translating scientific intent into executable automated workflows remains a bottleneck. We present a template-based platform that separates workflow design from execution. No-code interfaces combined with artificial intelligence (AI) agent-assisted code generation enable domain scientists to create reusable workflow templates, that is, parameterized specifications of a complete experiment or simulation, from device orchestration or model definition through analysis and reporting to data storage. Templates are generated once and thereafter executed deterministically, without large language model (LLM) involvement at run time. The platform integrates multi-device laboratory automation, mechanistic simulation and standardized data pipelines with electronic lab notebook (ELN) documentation. We demonstrate the approach on a chromatographic case study comprising automated isotherm determination for two proteins on a mixed-mode resin, mechanistic simulation of two elution strategies and experimental column validation, implemented entirely through wizard-created templates without manual scripting. ... mehrRegenerating every artifact 20 times, one-shot and without manual correction, showed that reliability depends on the task and on the LLM. With the LLM deployed in the case study (GPT-5.1), workflow-structural scripts and device protocols without substantial logic were correct in 87 to 100% of generations, whereas the most complex protocol reached 50%. In contrast, the more capable GPT-5.5 reached 100% on the same protocol. Generated templates therefore require verification, supported by explicit review and, where necessary, version-controlled revision. By making workflow design accessible without removing determinism from execution, the approach offers a route to laboratory automation that is reusable, auditable and extensible beyond the domain demonstrated here.