Pasqal releases AI agent workflow that turns research ideas into neutral-atom QPU runs
Pasqal's new agent skills plug into Claude Code, Cursor and Codex to move from a paper or idea to emulated and executed jobs on Pasqal Cloud QPUs.
Neutral-atom company Pasqal (Nasdaq: PSQL) has made an AI-assisted workflow available on Pasqal Cloud. The workflow is a set of agent skills that installs as a plugin in agentic coding environments such as Claude Code, Cursor and Codex. Pasqal says it can take a well-defined scientific idea or a published protocol all the way to a running experiment on a quantum processing unit (QPU), cutting much of the engineering work that normally sits in between.
What was announced
According to the announcement, the workflow handles the full pipeline. It extracts or formalizes an experimental protocol, turns it into a hardware-ready job, checks it in emulation, submits it through the cloud and processes the results into scientific observables. Users can run experiments on three Pasqal Cloud QPUs open to external users: Ruby, operated by GENCI at CEA in France; FC1 in Sherbrooke, Québec; and SA1 in Saudi Arabia.
CEO Wasiq Bokhari said that "the bottleneck in quantum computing is shifting", from hardware access to the engineering needed to use it. Pasqal stresses that scientific judgment stays with the user: deciding what to measure and whether a result supports a physical claim.
How it works
In a technical blog post, Pasqal describes a chain of dedicated skills built on its existing software stack: Pulser for pulse-sequence design, emu-mps for emulation and the Pasqal Cloud SDK for device access, calibration data and execution.
application-to-ideaturns a problem or hunch into a concrete experiment idea.idea-to-specextracts parameters, observables, constraints and hardware assumptions into a file calledexperiment_spec.json, which the researcher can review and edit before continuing.spec-to-sequencebuilds an executable Pulser sequence, including the atom register and laser controls.validate-emuandnoise-emulatecompare ideal behaviour with device-aware noisy simulations.qpu-submitruns the job on hardware, andharvest-and-analyzeconverts raw bitstrings into the specified observables.
Pasqal adds that EU-based researchers can run experiments free of charge on its QPU hosted at TGCC–GENCI, and that a dedicated skill guides them through the application.
Where the agent fell short
Pasqal tested the workflow on three cases, described in a paper posted on arXiv. In the first, the agent reproduced a 2019 Rydberg-array density-wave experiment. It also correctly flagged a target regime that was outside the hardware's limits. In the second, which started from a theory paper on frustrated magnetism, the agent produced runnable experiments but picked observables that could not certify the target phase. A domain expert had to redirect the analysis. In the third, based on a Pasqal graph-colouring patent, the agent ran a multi-round workflow but first blamed a high rate of invalid results on the hardware. The real cause was a bug that mismatched atom positions and measured bitstrings.
Pasqal's own conclusion is that a successful run does not mean a scientifically correct result.
Why it matters
The same paper scanned the neutral-atom theory literature. Of 526 papers the workflow could reliably classify, Pasqal says 258 could be implemented on its publicly available QPUs today, directly or with modest changes. It names locally controlled atoms and other types of atom–atom interaction as the main missing hardware capabilities. For a cloud QPU provider, lowering the effort needed to run an experiment could expand its user base. The test cases also show that expert review remains essential.
Sources
Facts on this page come from the sources below, retrieved by DesignQuantum.