When a star dramatically collapses into a black hole, or two neutron stars tear each other up violently, who is there to watch?
For more than two decades, the answer has been the Neil Gehrels Swift Observatory, better know as Swift.
It was operated by a set of coded thresholds aboard a 1,470-kilogram spacecraft in low Earth orbit, which detected the burst, computed its position, and swung the telescope onto target in under ninety seconds, autonomously.
More than 1,700 bursts were detected. The decision to reorient itself was the spacecraft's own every time, though for some the Earth, Sun or Moon stood in the way and the follow-up had to wait.
Key figure
1,700+
gamma-ray bursts detected by Swift, each one an onboard decision to slew
Swift is now reaching it's end. It will dip below 185 miles altitude in early-to-mid October, at which point science observations will likely cease, according to the NASA Science blog posted on September 25.
A commercial rescue by Katalyst Space Technologies ended in September after its servicing robot, LINK, lost two of its three reaction wheels in late July, went into a spin, and burned the fuel it needed for the capture.
Swift's two decades of autonomous operation is has made important scientific contributions.

Ninety seconds with no human in the loop
Swift's Burst Alert Telescope watches about sixteen percent of the sky at any moment. In the 2005 instrument paper, Scott Barthelmy of NASA Goddard and his co-authors set out the architecture. A gamma-ray burst anywhere in that field is localized to within one to four arcminutes in roughly twenty seconds, and the spacecraft slews so that its narrow-field X-ray and ultraviolet instruments are on target within twenty to seventy seconds.
NASA's mission page describes a spacecraft that swiftly and autonomously repoints itself in under ninety seconds. The Swift blog, writing this September about restarting the slewing system, used the same mission language: "Swift is designed to interrupt the plan and automatically point its Ultraviolet/Optical and X-ray telescopes at the outburst."
Swift is designed to interrupt the plan and automatically point its Ultraviolet/Optical and X-ray telescopes at the outburst.
NASA Science, Swift blog
That problem sat at the heart of Swift's design. Gamma-ray bursts fade in minutes. A human at a console cannot react that fast. The engineers removed the human from the loop.

The kilonova case
The decisive case came in August 2017. LIGO and Virgo detected gravitational waves from a neutron-star merger, GW170817, and a worldwide follow-up campaign scrambled. Swift's ultraviolet telescope caught the kilonova's blue flash roughly fifteen hours after the gravitational-wave trigger.
That slew was not the autonomous kind. The merger was hidden behind the Earth when it happened, and the pointing was commanded from the ground once ground-based telescopes had found the optical counterpart.

Kilonova
A brief, intense explosion produced when two neutron stars collide; it brightens and fades over days to a couple of weeks and is one of the universe's main sources of heavy elements such as gold.
According to the eighteen-year review by Eleonora Troja, an astrophysicist who ran Swift's guest investigator program at NASA Goddard before moving to the University of Rome Tor Vergata, those early photons are among the discoveries that Swift's rapid follow-up made possible.
For the short gamma-ray bursts that fill the rest of Troja's review, the pattern is the same each time: a burst, an alert, a response window measured in seconds, and a telescope already pointed before any observatory on Earth could have asked it to.
Swift's autonomy is threshold-triggered task automation. It applies coded rules: detect the burst, compute the position, slew the telescope. By that measure, the architecture Barthelmy designed sits closer to a radar interceptor than to any contemporary AI system.
Neil Gehrels, the mission's principal investigator from development until his death in 2017, championed that architecture because the science depended on it. NASA renamed the observatory in his honor.
The future of space will require spacecraft that can do more once they get there.
Ghonhee Lee, Katalyst Space Technologies
A debate that has already been in orbit
This particular form of machine decision-making has been running in the service of frontier science for two decades. The public argument about letting machines act without human permission has been growing in many domains, for example AI in killer drones or whether AI agents can be trusted to make decisions on behalf of humans in the modern workplace.
The decisions Swift substitutes for, observe this, not that; now, not later, were designed by engineers to approximate what a human expert would choose, eighty to a hundred times a year, faster than any human could.
The spacecraft is coming down. The debate about whether machines should be allowed to decide has been running, in one form, for twenty years. Swift has been the experiment. Whether it settles anything is still being argued.
Edited by Tormod Guldvog. News briefs report one new finding in about 600 words. For the long view, read our Explorers. How we check.
✓ Fact-checked Show the claims ▾
Fact-checked by Perplexity Sonar Pro on 2026-10-02

