Scientists nowadays spend a significant share of their time not only testing their hypotheses but also integrating and optimizing the tools that make experimentation possible. This optimization and control take up most of an experiment’s timeline. But with the emergence of AI, the way science gets done is changing. Initially, researchers used AI in the digital domain to support experiment planning, data analysis, and interpretation. Now, researchers are exploring how it can bridge the gap between digital decisions and physical execution by controlling instruments and coordinating workflows across the lab.
In practice, though, this is far from straightforward. Labs run on multiple specialized instruments. Each has its own software, operating states, safety limits, and physical constraints. A model may understand the goal, but it still needs reliable feedback on when an instrument is ready, which protocol it can run, whether the plate is in place, and when another device can move without causing a collision. Efforts like Anthropic’s recently previewed Model Hardware Standard (MHS) are beginning to solve this by giving AI agents a common way to discover and operate lab instruments. For any approach like this to work, instruments need accessible, programmable interfaces. Many lab instruments aren't built this way, they sit behind proprietary software and closed drivers, turning integration into a weeks-long effort.
Now imagine describing an experimental workflow in natural language:
“Dispense 20 µL of detection reagent across the 96-well plate using the Mantis, then transfer the plate to the plate reader using the robotic arm. Measure fluorescence at 450 nm and record the results. Identify positive wells (exceeding the fluorescence threshold of 0.5 RFU), then collect samples from these wells and transfer them to fresh tubes using the FLO i8 PD.”
This workflow is illustrative. Every step depends on instruments that can be queried and controlled programmatically.
At Formulatrix, we've always built differently. All instruments in our portfolio ship with fully documented, open APIs. No black boxes. No vendor lock-in. Just direct access to the instrument.
That design philosophy has always made our instruments easier to integrate into custom workflows and automated pipelines. Now, as AI-driven lab automation moves from concept to reality, it's making them a natural fit for the next generation of intelligent labs.
Open APIs In Practice
Every Formulatrix liquid handler comes with a documented API. That means any external software (a custom scheduler, a LIMS, a robotic arm controller, or an AI agent) can do the same things a researcher would do manually: start a run, check progress, adjust parameters, and know when a step is complete.
This matters because real lab workflows cross instrument boundaries. A dispense step feeds into a downstream assay. Adding a reagent triggers a plate handoff. A screening run determines whether the next round happens today or next week. Instruments that can't communicate with the rest of the workflow create bottlenecks that cost researchers time, samples, and sometimes entire experimental runs.
Our liquid handlers are designed to fit into those workflows, not stand apart from them.
What This Looks Like at the Bench
The University of Washington case study is just one example of what AI-driven lab automation looks like in practice. As agents supervise longer, more complex experimental workflows, the demands on instruments rise accordingly. Adjusting protocols in real time, chaining operations across instruments overnight, flagging failures before hours of work are lost: these are the workflows coming to academic and industrial labs alike. The liquid handlers at the center of these workflows will be the ones built openly.
Formulatrix Open Ecosystem - Liquid Handler Integration via Open API
A Portfolio Engineered for Integration
Mantis®: Non-contact, chip-based dispensing from nanoliter to microliter volumes. Its microfluidic chip technology handles viscous liquids, beads, and cells with a CV of less than 2% at 0.1 µL, making it the strong choice for sensitive, low-volume workflows where consistency across every well matters.
Tempest®: Designed for high-throughput workflows that can't afford slowdowns. With 96 output nozzles, support for up to 12 reagents, and a full 96-well plate dispensed in only a few seconds, it keeps pace with the most demanding screening campaigns without sacrificing accuracy.
F.A.S.T.™: A 96-channel positive displacement liquid handler built for high-throughput assay workflows. Positive displacement makes it liquid-class agnostic, handling everything from aqueous buffers to viscous or sensitive samples, with a flexible tip configuration (1, 8, 12, or 96 tips) and computer vision tip detection.
FLO i8® PD: An 8-channel general-purpose liquid handler with fully independent channels across span, Z-axis, and volume. Handles 0.5 to 1000 µL with contact dispensing, and down to 200 nL with non-contact dispensing (NCD). Dynamic liquid-level detection and touch sensors automatically map well geometry, reducing setup time and user error. Optional UV decontamination and HEPA filtration modules expand the system's capabilities for workflows requiring enhanced contamination control.
Reagent Exchanger: 96 aspiration nozzles, up to 6 dispense sources, plate tilting, automated delidding, and automated cleaning, all in a compact footprint. Built to keep high-throughput media exchange and plate washing running smoothly with minimal manual intervention.
Mantis - Precise Liquid Dispenser
Tempest - High-Throughput Dispenser
F.A.S.T. - Positive Displacement Sample Transfer
Flo i8 PD - General Purpose Liquid Handler
Reagent Exchanger - Automated Plate Washer
The Lab of the Future Runs on Open Infrastructure
Conventional lab automation was built for scale, not for change. It pays off when a single protocol runs thousands of times, which is why it has always fit manufacturing and high-throughput screening better than research. A lab that runs dozens of protocols a year, half of them new, cannot justify the integration cost, and most academic labs never have.
AI agents change that equation, but only if the instruments can communicate openly. An agent that can't query an instrument, can't receive its output, or has to navigate proprietary software just to start a run isn't a practical tool; it's a demo.
Formulatrix liquid handling instruments are ready for the workflows that are coming: not because we retrofitted AI compatibility, but because we built them openly in the first place.
Related reading: Anthropic — Previewing the Model Hardware Standard