REEF Imaging Farm

REEF is a self-driving cell-biology lab. It is an automated imaging farm where AI agents plan, run and adapt live-cell experiments across a whole cluster of instruments, with a human able to watch and step in at any point.

The idea at its heart is simple: a scientist should be able to describe an experiment in plain language and have the lab carry it out. Under the hood, REEF brings together multiple microscopes and fluidic systems, robotic arms, liquid-handling robots and automated incubators into one coordinated system. AI analyzes the images as they are captured and feeds decisions back to the hardware, so an experiment can adapt while it runs, minimize phototoxicity, and catch rare events in living cells.

One cycle, end to end

REEF runs the same loop around the clock. This is one turn of it, rendered from the lab’s digital twin: a plate leaves the incubator, is imaged on a microscope, travels to the liquid handler for a media exchange and a reagent addition, and goes back to the incubator until the next timepoint.

The geometry is the lab’s own. Every instrument stands where it stands, the liquid handler’s deck is its real layout down to the plate positions, and the robot follows the paths it is actually taught. The pipetting is drawn simplified, for legibility rather than as a simulation.

The farm in action

And here is the same workflow on the real hardware. The robotic arm moving a plate between the incubator and the microscope, part of a fully automated imaging run:

And the whole farm during a run, across its lab cameras (transport, imaging and liquid handling at once):

What it is for

• Automated widefield and fluorescence imaging at scale
• Long-term live-cell imaging and cell tracking
• Spatial-omics and multiplexed imaging
• Real-time, AI-augmented analysis with feedback control, automatically adjusting field of view, illumination and other conditions on the fly

The near-term goal is massive, high-quality image-data generation. The longer-term goal is experiments that a scientist can simply talk to.

A live, agent-run experiment (2026)

In 2026, REEF ran a live, fully agent-controlled wet-lab experiment, demonstrated on stage during an invited talk. A researcher typed a single natural-language request, to study how osmotic pressure affects living cells and then attempt a rescue to see if the changes reverse, and an AI agent orchestrated the whole lab remotely, streaming images to a live viewer in real time. No manual pipetting, no clicking through software: one prompt, and the farm ran the experiment on real cells, end to end, catching and recovering from the inevitable live hiccups along the way. It worked, and the osmotic change was confirmed reversible.

We wrote up the experience, including a short interview with the agent that ran it, here: One Prompt, a Whole Lab: A Live Agent-Run Experiment on REEF.

That run was made possible by Hanzhao Zhang (experiment preparation and getting the system ready for full runs), Songtao Cheng (hardware, and keeping everything running through the live demo) and Wei Ouyang (who took it onto the stage), with contributions from the whole AICell Lab team.

Built around safety

REEF pairs a capable AI agent with a safety-first system. Experiments stream to a live viewer, a human stays in the loop, and unsafe actions are refused rather than risked, so the cells and the instruments are protected even when a run hits the unexpected. (We keep the deeper technical details under wraps while the work is unpublished.)

Get involved

REEF is where AI agents, robotics and live cell biology meet. The code lives on GitHub, and if building labs that a scientist can simply talk to sounds like your thing, we would be glad to hear from you.