The Self-Driving Lab Is Official: SLAC Is Handing Its X-Ray Laser to an AI Copilot

The world's most powerful X-ray laser has a problem no staffing plan can fix. It generates data faster than any human team could ever read it. Last week, the Department of Energy decided the solution is not more scientists. It is a machine that does the science itself.

On October 8, at the Golden Age of Science Summit in Washington, D.C., the DOE's Office of Science announced four national laboratory-led projects under its Robotics and Automation Testbeds for Autonomous Scientific Discovery program, a $30 million initiative that treats the self-driving laboratory as national infrastructure rather than a curiosity. The flagship of the four is SPIRE, a Source-to-Discovery Platform for Instrumentation, Robotics, and Embodied AI in DOE Photon-Science Facilities, led by SLAC National Accelerator Laboratory in California.

The firehose a million pulses wide

To understand why a national lab is handing its instruments to AI agents, start with the instrument. SLAC's Linac Coherent Light Source, the LCLS, was the world's first X-ray free-electron laser, capable of imaging proteins atom by atom and filming chemical reactions at femtosecond timescales. Its successor, LCLS-II, runs on a superconducting accelerator that fires up to one million X-ray pulses per second. When the detectors operate at full power, they will produce data at rates exceeding one terabyte per second, roughly a thousand full-length movies every second, or the contents of a major research library streaming through the instrument in minutes. No shift schedule, however generous, can analyze a firehose that wide in real time.

This is the central mismatch of modern photon science. Experiments have become exascale problems in miniature: samples measured in milliseconds, beams adjusted on timescales ranging from microseconds to hours, decisions compounding faster than a human operator can follow. The bottleneck is no longer the brightness of the beam or the sensitivity of the detector. It is the speed of comprehension. SPIRE is the institutional response, not an incremental upgrade to an existing process, but a structural redesign of the path from experiment to discovery.

What the robot scientist actually does

SPIRE will tie together the physical realities of the beamline: robotic sample manipulation, intelligent detectors, accelerator and instrument controls, and a persistent record of the experiment's state, all connected through edge-to-high-performance-computing pipelines into a single autonomous workflow. The initial deployment covers SLAC's two flagship user facilities, the LCLS and the Stanford Synchrotron Radiation Lightsource, or SSRL.

The workflow reads like a laboratory run by an attentive colleague who never sleeps. A super-intelligence agent, SI in the DOE's current terminology, interprets incoming data as it arrives and adjusts the experiment on the fly, directing robotic arms to make delicate changes to samples and X-ray beams. (Autonomous agents are spreading beyond the lab too inside the enterprise AI agent wave.) Instead of archiving everything and sifting later, the system filters for only the most valuable results and streams them to researchers, who walk away with immediate insights instead of a mountain of raw data to process. The experiments themselves cover nearly every kind of sample imaginable, from fusion fuel targets to water molecules, plant stems to proteins, sturdy battery cells to ancient manuscripts.

The science already runs faster than the scientists. The lab of the future is the one that stops waiting for humans to catch up.

One of four, and designed to travel

SPIRE does not stand alone. The four funded projects map the full lifecycle of autonomous science. MAESTRO develops capabilities that learn and improve through experience. DART stress-tests those capabilities to determine whether they transfer reliably. TRACE, led by Oak Ridge National Laboratory, packages and connects them through standardized interfaces, typed capability contracts, and digital twins, so autonomy can be tested, reproduced, and deployed across laboratories without rebuilding the complete software stack. SPIRE then deploys and validates that autonomy in the consequential, real-world operations of photon and electron science facilities.

Crucially, SPIRE is being built as a testbed meant to translate. SLAC is partnering with Stanford University, the University of Chicago, Argonne National Laboratory, Brookhaven National Laboratory, and Lawrence Berkeley National Laboratory, plus advisors from industry, to design methods, interfaces, digital twins, and benchmark tasks that other DOE user facilities can adapt. A system that works at one beamline should not stay at one beamline. Total funding across the four projects is $30 million, with $2 million in fiscal year 2026 and outyear funding contingent on congressional appropriations. Selection for award negotiations is not yet a final funding commitment, and the DOE can cancel during negotiations.

The Genesis Mission puts AI inside the experiment

The announcement landed alongside a much larger commitment. The same week, the DOE unveiled $159 million for 12 new Genesis Mission Phase II projects, the national initiative to mobilize AI for science and technology, bringing the mission's first-year portfolio to 297 projects. The selections span a fusion-device digital twin led by Commonwealth Fusion Systems for its SPARC demonstrator, RNA-structure research, application-aware quantum error correction, geothermal reservoir modeling, rare-earth separation, scientific software, and laboratory operations. NVIDIA separately pledged a $1 billion, five-year commitment of compute support for Genesis Mission science, announced at the same summit.

SLAC itself is carrying more than SPIRE. The lab was also selected to lead a super-intelligence-driven autonomous discovery project in catalysis, a closed-loop system that combines real-time X-ray measurements with scientific evidence and hypothesis ranking to choose the next experiment, accelerating mechanistic discovery in processes central to energy and chemical production. It will partner on eight other Genesis Mission projects spanning quantum materials, microelectronics, biosciences, and fusion energy. DOE Under Secretary for Science Dr. Dar\u00edo Gil called the Phase II awards a critical step in turning the promise of AI for science into transformative capability.

What changes when the instrument reads itself

The deeper shift is about who decides what an experiment means. In the old workflow, a scientist booked beamtime, collected data, and spent weeks or months deciding whether the run was worth repeating. In the autonomous workflow, that judgment is made continuously, by software with access to the beamline itself. SPIRE lead PI Angelo Dragone, deputy associate lab director for SLAC's Technology Innovation Directorate, frames this as a change in the tempo of discovery: the autonomous laboratory design, he said, opens up entirely new ways of making discoveries while dramatically shrinking the time between running an experiment and understanding it.

There are honest questions beneath the optimism. Autonomous systems inherit the biases of their training and their reward functions, and a benchmark task is only as trustworthy as the metric behind it. The TRACE project's emphasis on independently rerunnable evaluation and end-to-end provenance is a recognition that reproducibility has to be engineered into self-driving science, not assumed. And the DOE's own framing of the $159 million Genesis awards, which bundles work as different as chip design and RNA databases under one banner, means the real measure of success will be scientific results per project, not announcements per press release.

Still, the direction of travel is unmistakable. When the instrument produces a terabyte a second, the limiting factor in discovery is no longer the beam. It is the reader. SLAC's bet is that the reader can be built, and that the lab of the future is not a bigger building, but a tighter loop between measurement, reasoning, and the next question.