Danaher and Microsoft are developing an AI platform designed to help scientists plan and run experiments using laboratory instruments. In preliminary testing described by Unite.AI, novice scientists using an assistant completed an experiment up to 83% faster, though the result comes from Danaher’s internal testing, not an independent study.
Microsoft AI Watch analysis
What happened
The companies announced a strategic partnership on 20 November to build a shared agentic platform for scientific instruments. It is intended to interpret results and guide instrument control, using Azure and Microsoft Foundry. Danaher says the platform should be reusable across its products and open to multi-vendor laboratories.
The first named application is MetaXpress Acquire AI Assistant, software for Molecular Devices’ high-content cellular imaging systems, designed to provide product support. The broader platform is intended to help translate scientific goals into experimental plans, set up instruments, guide execution and troubleshooting, and interpret results. The companies say instruments would operate within defined safety boundaries, with scientists reviewing recommended corrections and next experiments. Initial commercial applications are expected in 2026, with more agent-enabled instruments planned for 2027.
Danaher reported that, in preliminary internal testing on a Leica microscope, novices using AI assistance took up to 83% less time and 68% fewer steps than novices working without it. The assisted novices completed the experiment in a time comparable to an expert. Those figures describe different comparisons: the reductions are against novices without assistance, while the expert comparison concerns completion time.
Why it matters
This is a concrete attempt to move AI from answering questions about science to helping operate the equipment that produces scientific results. If the approach works across different instruments and laboratory suppliers, it could make experimental workflows easier to reuse and connect. The early figures are promising, but internal preliminary testing is a starting point, not a settled measure of performance.
Our read
The interesting part is the proposed combination of instrument control and scientist oversight, not the word “agentic” doing its usual heavy lifting. A system that can help plan and run experiments may save time, but the useful test will be whether other laboratories can reproduce the results while keeping control of consequential decisions.
What to watch
- Whether the first commercial applications arrive on the stated timetable.
- Which instruments and suppliers the shared platform supports.
- Whether Danaher publishes more detail on testing methods and results.
- How safety boundaries and scientist review work in practice.
Discussion spark: Should AI systems be allowed to control laboratory instruments within defined limits, or should a scientist approve every experimental step?
Sources and evidence
- Danaher Partners with Microsoft on Agentic Lab Instrument Platform – Unite.AI (20 November 2026, 12:50 UTC)
not affiliated with or endorsed by Microsoft