A Shared Language for Lab Hardware Opens the Door to Scheduling and Record Services

A test of turning each instrument’s language into one

Anthropic released a limited research preview of the Model Hardware Standard (MHS) on August 27, 2026. It is a connection protocol intended to let AI systems that carry out multi-step tasks find and handle different instruments—such as microscopes, liquid-handling equipment, and robotic arms—in a consistent way.

An instrument’s connection file records the status it can report, the commands it can run, the values that can be changed, and the safety limits that must not be crossed. The standard applies to equipment that can receive commands from other software. Closed instruments that can only be operated through screen buttons require additional work from their manufacturers.

In an early trial, Genentech connected liquid-handling equipment, a robotic arm, and three measurement instruments for protein analysis. One microscope system at HHMI Janelia Research Campus reported reducing a process that launched software from seven companies in sequence to one click from a single screen. These are early reports from Anthropic and participating institutions, not evidence that ordinary laboratories will see the same results.

The immediate issue may be operational burden rather than connection technology. According to a 2025 report from the National Research Facilities & Equipment Center, which supports the management of national research facilities and equipment in Korea, university equipment had an average utilization rate of 26.7% in 2024: roughly one hour of actual use for every four hours it was available. There were 0.45 dedicated staff members per instrument, meaning fewer than one person was responsible for two instruments.

If this shift fails, a likely reason is building instrument-control features first without identifying where booking omissions and record mismatches occur. MHS is still a research preview accepting applications, with a goal of later public release after safety evaluation.

A day for one person managing 12 instruments

Consider Assistant Manager Kim, a hypothetical operator managing 12 instruments alone in a university shared-equipment facility. Each morning, he checks the booking screen, finds training-completion records in a separate file, and asks the researcher in charge whether an instrument is under repair.

When a researcher needs to use liquid-handling equipment and a measurement instrument in sequence, they must manually add preparation time between the two bookings. If the earlier experiment runs late, the following reservation does not shift automatically. The operator has to coordinate the time again by phone or email.

After use ends, the operator compares the booking record, paper usage log, files produced by the instrument, and usage-fee billing details. Korean national guidelines also require online booking, operating logs, and shared-use performance management for shared equipment, so if even one item is missing, staff must track people down again at month-end.

Once a common connection layer is in place, one screen could show today’s bookings, current instrument status, and whether the previous task has ended. When liquid-handling equipment sends a completed status, the system could calculate preparation time for the next measurement instrument and show the operator only proposed changes when there is a conflict.

AI does not need to control instruments directly from the start. Simply reading an instrument’s status file and booking records to confirm that use has ended, then moving to the next step after human approval, can reduce manual re-entry.

Some work must clearly remain with people. An on-site person in charge must judge whether a sample has bubbles, whether someone is near the instrument, and whether an expensive sample can be used again. In the Genentech trial, AI initially treated bubbles as if they were a software error and needed human guidance.

The completion criterion for a first product, then, is not autonomous instrument operation. A more suitable goal is one screen showing whether a recent booking connected without gaps to actual use, records, and billing—and preventing incorrect commands from being sent at all.

How far shared research facilities have gone

Northwestern University in the United States used NUcore, a shared research facility management programme, to combine bookings, service orders, usage-fee billing, and usage reporting. Each facility charges users for instrument time and services, while the NUcore programme itself was made available for other institutions to adopt at no cost.

In 2019, the university’s full network of shared research facilities participated. In fiscal year 2020, it handled more than 4,000 active users and USD 19.7 million in billing. It is an example of scaling by standardising bookings and money flows before instrument control.

HHMI Janelia Research Campus in the United States began expressing the status of lasers, cameras, and sensors in a common format through a custom microscope that became the starting point for MHS. It later expanded the work to a fish-brain imaging system that used software from seven companies and to other microscope experiments.

The institution reported that adding one laser-monitoring camera fell from several days to a few minutes. This was an internal research trial: no separate usage fee was disclosed, and it is not a commercial service that operates all microscopes without human supervision.

Stanford University Department of Pediatrics in the United States applied Calira to equipment bookings used by more than 250 researchers across 51 laboratories in five buildings. Starting with scattered calendars and notes about equipment failures, it brought together booking, maintenance-time blocks, user notifications, manuals, and usage agreements.

It now also uses recorded usage time to allocate costs among laboratories. Specific equipment fees and software contract costs were not disclosed, and the results come from a supplier-published customer case study, so independent verification is needed.

Four services to start small

1. Sequential equipment booking assistant

• What it does: Books several instruments needed for one experiment at once, including sequence and preparation time.
• User: A graduate student who needs to use liquid-handling equipment and a measurement instrument in sequence on the same day.
• Why now: The possibility of reading instrument status in a common format means delays in an earlier task could be reflected in the next booking.
• First screen: Below the experiment start time, show the equipment sequence, open slots, preparation time, and reasons a booking is unavailable side by side.

2. Experiment sequence check assistant

• What it does: Does not move instruments. It compares current status with the experiment sequence and flags only missing steps or risky progress.
• User: A new researcher using automated liquid-handling equipment for the first time who cannot call an operator every time.
• Why now: AI’s physical judgment is still incomplete, so using it for checking and approval is less risky than direct control.
• First screen: Show today’s step list, evidence of completion, the next step, and an operator approval button.

3. Usage-record reconciliation checker

• What it does: Compares bookings, actual operating time, usage logs, and usage-fee billing to find items that do not match.
• User: An operator at a university shared-equipment facility who reconciles shared-equipment performance and billing at month-end.
• Why now: Korean guidelines require several records, while dedicated staffing is below one person for every two instruments.
• First screen: Collect only bookings with mismatches and show the missing record and the person who needs to confirm it.

4. Equipment connectivity assessment service

• What it does: Checks whether existing instruments support status queries and remote commands, then sets a connection order.
• User: A university laboratory lead or equipment distributor using older instruments alongside new robots.
• Why now: MHS starts with instruments that can receive software commands, so connectivity needs to be distinguished before purchase.
• First screen: After entering a manufacturer and model name, show three levels: status query available, manufacturer discussion needed, and difficult to connect.

One thing to check today

Call an operator at a shared-equipment facility and spend 30 minutes tracing how three recently completed bookings moved from reservation to usage log and billing. If two or more of the three required the same information to be entered twice, or took more than 10 minutes to confirm an omission, it may be worth testing a service that connects records before attempting instrument control.

Why this matters where you are

Shared research facilities in your market may have different rules, billing arrangements, and staffing levels, but the operational question is portable: do bookings, actual usage, logs, and charges connect without manual reconciliation? Trace a few completed bookings before designing automation. The first useful product may be a record and approval layer rather than a system that moves physical equipment.