Quantum computing experiments are slow and measurement-heavy: preparing and running qubit experiments can take months and require hundreds to thousands of preliminary measurements. Researchers at MIT’s Engineering Quantum Systems Group (EQuS) have been exploring whether AI agents can streamline that work. Graduate student Beatriz Yankelevich used GPT-5.6 Sol, harnessed to Codex, connected directly to the lab software that coordinates experiments, allowing the agent to run measurements, analyze results, and decide what to try next.

The group studies superconducting qubits, which are cooled to near absolute zero inside dilution refrigerators, controlled with microwave signals, and arranged on a chip. Once a chip is fabricated, packaged, and cooled, researchers interact with it entirely through software, making the experiments a natural testbed for AI agents. Superconducting qubits are often called artificial atoms because they occupy only specific energy levels; microwave pulses move qubits between these levels and probe their quantum state. Researchers design and calibrate these pulse sequences, then digitize and analyze the returning signals to find resonance frequencies, coherence times, and the settings needed for computation.

Calibrating qubits requires a series of interdependent measurements, with each result shaping what comes next. Qubit properties can drift, and unexpected physical behavior can cause inconsistent results, so experienced researchers must recognize and adapt to changes. That combination of software control, repeated measurement, and adaptive decision-making makes qubit calibration a compelling use case for AI agents.

Yankelevich tested GPT-5.6 Sol on an uncalibrated six-qubit chip of a standard type that EQuS routinely uses to benchmark fabrication. She provided Codex with measurement-specific skills explaining how to run and evaluate every experiment. Using those skills and the chip’s design targets, the agent selected measurement parameters, operated the hardware, analyzed the resulting data, and then either refined the measurement or saved the result for the next step. When signals were clear, Codex completed a standard measurement sequence with little intervention: it identified transition frequencies, calibrated control and readout pulses, and determined how long the qubit retained quantum information.

When signals were weak or noisy, GPT-5.6 Sol took longer to find suitable parameters and sometimes needed guidance from an experienced researcher. The results suggest current agents can handle clearly defined workflows, but interpreting ambiguous physical results remains a challenge. EQuS now regularly uses agents to handle routine chip characterization, which can otherwise take several days per chip. Yankelevich reported running agents overnight or while working in the cleanroom, checking in from her phone to steer them when needed. For novel experiments, she assigns agents narrower goals and relies more on their ability to write, modify, and test new code against real measurements. She has built infrastructure guiding agents across measurement, theory, and chip design, allowing multiple agents to work in parallel while she focuses on higher-level work such as interpreting results, devising experiments, and planning next steps.