Chapter 03 — One year, autonomous labs

Colleagues —

On November 24, 2025, the White House issued Executive Order 14363, Launching the Genesis Mission. One year is a useful interval because the order itself built in clocks. I am not going to invent a birthday narrative. I am going to read those clocks against the labs that were already building self-driving experiment loops before anyone named the program.

I am using the same anniversary standard on the Factory. A virtual assistant that drafts a cute reply and never books the showing is a demo, not a product. Closed loop: missed-call, text-back, booking — overnight.

What the order actually required

EO 14363 is not vague about infrastructure. Section 3 stands up the American Science and Security Platform and tells the Secretary of Energy to provide, in an integrated way:

HPC, including lab supercomputers and secure cloud AI environments
AI modeling and analysis frameworks, including AI agents to explore design spaces, evaluate experiments, and automate workflows
domain-specific foundation models
secure access to federal, proprietary, open, and synthetic datasets
experimental and production tools for autonomous and AI-augmented experimentation and manufacturing

Then the deadlines:

90 days — identify federal computing, storage, and networking available to the mission.
120 days — identify initial data and model assets; plan to ingest other-agency and partner data.
240 days — review capabilities across DOE labs and other federal facilities for robotic laboratories and production facilities that can do AI-directed experimentation and manufacturing, including standards.
270 days — seek to demonstrate initial operating capability of the Platform for at least one national challenge.
One year, then annually — the Secretary reports to the President, through OSTP and OMB, on platform status, lab integration, user engagement (including students), research outcomes, partnerships, and gaps.

If you are inside the complex, the question for this anniversary is simple: did the 240-day robotic-lab review produce a real inventory, and did the 270-day IOC demo happen for a named challenge? The first annual report is due this month. That document — not a summit montage — is the artifact to read when it lands.

Autonomous laboratories, in DOE's own words

The July 2026 challenges document, "Achieving AI-Driven Autonomous Laboratories," states the bottleneck without romance: human-driven experimental loops and a shortage of non-deterministic AI control tools in combinatorially large design spaces. The prescribed stack is robotics, edge AI, real-time analysis, intelligent feedback, hypothesis generation, and data curation/sharing. Lead agencies: DOE, HHS, NSF, NIST. The White House July 22 release uses the same sentence: labs "which scientists can run and replicate from anywhere in the country."

That last clause is the AmSC problem again. A robot in Oak Ridge that cannot be orchestrated from a PI at a university is a demo, not a national capability.

The same document's materials challenge is explicit about closed-loop work: couple prediction, synthesis, characterization, and analysis so the system learns. Inverse design — specify the property, get the material — is named as the goal. Autonomous labs are how you close that loop without a postdoc running the same plate overnight.

INTERSECT was doing this before it had a mission name

Oak Ridge's INTERSECT initiative launched in 2022: interconnected "Smart Labs of the Future" for self-driving experiments, multi-modal measurements, and real-time AI/ML plus edge feedback. The architecture they publish is the one Genesis will have to scale — a human-AI-machine interface, experiment-management software, autonomous science applications, and a shared data layer. The autonomous chemistry lab on the main campus already has a mobile robot shuttling between workstations that have robots of their own. Sheng Dai's line is the one I keep: robots can work 24/7, and in the dark.

ORNL now frames INTERSECT as the foundation under Labs of the Future and under Genesis. The Autonomous Science page is the cleanest public description of the design–build–test–learn cycle they want: AI proposes conditions, automated instruments synthesize or prepare, characterization measures, ML picks the next experiment, scientists interpret and set priorities. They also say the quiet part: cybersecurity, data governance, and valid results are first-class requirements, not a later patch.

Mariam Kiran's ORNL-led Genesis Phase I project — "AI-powered operations for self-driving science" — is the networking version of the same idea: coordinate agentic and scientific workflows across institutional boundaries. That is INTERSECT leaving campus.

Fourteen robotics awards, and the facilities that will eat them

On December 10, 2025, DOE announced over $320 million to stand up Genesis AI capabilities. Buried in that package: 14 projects in robotics, automated laboratories, and autonomous control of large-scale experiments, aimed at "embodied AI, advanced automation, and robotics." The same announcement funded AmSC, ModCon, and 37 foundational-AI awards to curate data and refine models that can "drive autonomous laboratories and design new experiments and instruments."

That is the intended sandwich: models that propose, robots that execute, facilities that generate the next training set.

Berkeley Lab's AmSC Q&A names the facility-scale versions:

SYNAPS-I (SYnergistic Neutron and Photon Science – Intelligence) — embed AI into particle accelerators and synchrotrons; compress analysis from months toward near real time; make synchrotron and neutron user facilities behave more like self-driving labs.
OPAL — Orchestrated Platform for Autonomous Laboratories to Accelerate AI-Driven BioDesign.
MOAT — Multi-Office particle Accelerator Team.

Fermilab's Genesis award is the accelerator-control version: AI/ML resonance-control algorithms for superconducting RF cavities on PIP-II and a list of peer machines (LCLS-SC, EIC, FRIB, ATLAS). Brookhaven is building self-learning digital twins for accelerator operations and an AI-controlled cloud chamber. These are not "AI for science" posters. They are closed loops on instruments that already have queues measured in months.

A commercial door, if you build robots

The FY26 SBIR/STTR Genesis Phase I round — pitches were due September 10 — listed Achieving AI-Driven Autonomous Laboratories as one of four topics. Areas of interest on the ConnectWerx page: advanced robotics, AI for network operations, diagnostics, remote handling, and neuromorphic computing circuit primitives. A broader Phase I was promised for fall 2026. If you missed September, that fall round is the item to watch rather than inventing a new RFA.

DOE's collaboration page also lists Robotics and Automation as one of five consortium working-group themes, alongside models, data standards, HPC/cloud, and workforce. Partnerships still route through genesismission-partnerships@hq.doe.gov.

The AI Automation Factory

close the loop overnight

EO 14363 told Genesis to judge itself by clocks. I am using the same standard on the Factory.

A robot in Oak Ridge that cannot be orchestrated from a university PI is a demo, not a national capability. A "virtual assistant" that drafts a cute reply and never books the showing is the same demo. Closed loop for a lab: prediction → synthesis → characterization → next experiment, overnight. Closed loop for a broker: missed-call → text-back → booking → CRM note, before the lead goes cold.

Sudipa is the first bench. If the VA cannot run the overnight part of her desk — the after-hours inquiry, the same-day confirmation — it is not a product. Models without overnight experiments are incomplete. Assistants without overnight follow-up are incomplete the same way.

Clock: first paid client / cash by 31 December 2026. $2,500 setup + $2,000/month. December is the one-year mark for the order and the cash mark for the shop.

Worth clicking

Launching the Genesis Mission — reread Section 3(e)–(f) and Section 6. The clocks are in the order.
Autonomous Science | ORNL — the most honest public description of closed-loop labs in the complex.
INTERSECT | ORNL — the 2022 software/robotics substrate Genesis is inheriting.
Energy Department Advances Investments in AI for Science — the December 2025 $320 million package, including the 14 robotics projects.
Office of Science User Facilities — 28 facilities, tens of thousands of users, peer-reviewed access. Autonomous control only matters if it changes their beamtime.

December: agents. Not the consumer kind. The ones DOE is trying to make hypothesise, call a simulation, and queue an experiment without a new Slack thread.

The AI Automation Factory. Virtual assistants for sales and broker desks. First workflow: a real-estate broker/operator (Sudipa). Offer: $2,500 setup + $2,000/month. First paid client / cash by 31 Dec 2026. 2027: this newsletter + one weekly ask. No spray.

Reply if one loop on your desk is eating the week. deep-heap.com

Deep Datta
Deep Heap

Intelligence has a substrate.

I write this for people who already know the labs, the machines, and the constraints — and for operators who need a virtual assistant that actually runs a desk. If I cannot verify a source, I leave it out.