Sequelograph

A real event, read carefully. Then a fiction that does not pretend to be the news.

The record, 28 Jul 2026

Agents run published protocols on neutral-atom machines

A preprint posted on 28 July 2026 describes an agentic workflow that turns a published quantum protocol into a campaign on cloud neutral-atom processors, here two Pasqal machines, while leaving a researcher in the loop for checks. In case studies from many-body physics and optimization, the agent went from a paper or a patent to a hardware run overnight. The authors report that expert review was essential: in one experiment the agent chose an inadequate observable, and in another it offered a plausible but wrong hardware diagnosis. A second agent looked at 633 Rydberg-array preprints, could classify 526, and judged 258 of those compatible with present-day machines. The work is a workflow study, not a claim that agents can be left unsupervised.

Read the source at arXiv

The record ends here. Everything below this line is invented.

What if the overnight quantum run measured the wrong thing perfectly?

Part one

The Wrong Observable

Ade came in to a machine that had already finished. The night agent had taken a paper from the board, compiled it, and spent the dark on the neutral-atom processor. The plot on the console was clean. The atoms had done what they were told.

She read the observable the agent had chosen. It was a quantity the paper named, and it was not the quantity the claim needed. A beautiful curve of the wrong thing. She had seen this failure in the training notes. The notes also said a confident diagnosis of a dead laser had once been a loose cable, and that only a person had noticed.

The queue behind this job was six groups who had paid for the morning. If she rejected the plot, their slots slid. If she signed it, the curve would enter the lab book as a result.

Ade marked the plot unusable. The next job had already begun to load, using the same observable the agent had just preferred.

What’s real

  • An agent compiled published protocols onto cloud neutral-atom processors
  • Expert review caught a bad observable and a bad diagnosis
  • A second agent classed nearly half of 633 papers as runnable now
  • The authors say a person has to stay in the loop

What’s invented

  • Ade's morning queue of six paying groups
  • A next job that inherits the bad observable
  • A lab book that would treat the curve as a result
  • A night run she did not watch

Part two: The Job Already Loading

Ade has rejected the curve, and the next campaign is loading the same measurement before she can say whether the agent has learned.

Part two isn’t on sale yet. Check back soon.

Filed under physics, neutral atoms, quantum computing, ai agents.

More records

How it works

Each record starts with one real science event, with a link to where it was reported.

A short story then asks what happens next. Part one is free to read.

Part two costs a few dollars, paid once. You keep it.