The Invention Layer: A Public-Record Conversation with Jay Gambetta on Genesis Mission, Agentic Discovery, and Quantum-Centric Supercomputing

When the U.S. Department of Energy announced the first projects selected under its Genesis Mission Request for Applications, IBM landed in a role that cuts across hardware roadmaps, national-lab partnerships, and a bet that agentic AI can reshape how scientists find quantum use cases—not the other way around. In public remarks tied to the announcement, Jay Gambetta, director of IBM Research and IBM Fellow, framed the mission as an invention problem spanning every layer of the stack.

This piece reconstructs Gambetta's on-the-record statements from IBM Research's Genesis Mission blog post. It is not a private interview; every quotation below is attributed to published sources.

"Every layer of computation"

Genesis Mission is the DOE's push to build an integrated science-discovery platform weaving together AI, supercomputing, quantum computing, and advanced instruments. IBM was selected to lead a Phase I project under the mission's RFA, and the company simultaneously pledged up to $50 million in quantum system access—powered by its 156-qubit Heron and 120-qubit Nighthawk processors—for DOE national labs and collaborators over five years.

Gambetta's opening line on the opportunity is blunt about scope:

"Achieving the ambitious vision of the Genesis Mission will require invention and innovation across every layer of computation — from hardware and architecture to algorithms."

That sentence refuses a narrow reading of "AI for science" as fine-tuning a chatbot on PubMed abstracts. IBM's Genesis project explicitly inverts the conventional workflow: instead of starting with a scientific problem and hunting for a quantum algorithm, researchers begin with proven quantum algorithms and deploy an agentic assistant to search literature for real-world problems that fit—checking candidates against formal criteria and explaining reasoning for human review.

Humans define the criteria and validate proposals; the agent performs the search volume no single researcher could sustain.

Quantum-centric supercomputing in practice

Gambetta situates Genesis inside IBM's longer "quantum-centric supercomputing" vision—HPC, quantum, and AI operating in concert rather than as isolated procurement lines. He pointed to a recent Oak Ridge National Laboratory collaboration simulating molten salts for fusion tritium production as an early template: AI agents screened candidate salts from a 70-year ORNL database, GPU supercomputers modeled them, and quantum computers tackled the hardest simulation details.

IBM's blog cites 15 quantum computers online, 97%+ average uptime, 250,000+ users, and Nighthawk systems with 5,000+ QuOps and throughput up to 100,000 circuits per second.

Why agents belong in a quantum mission

The agentic research assistant must match proven quantum algorithms to problem classes, not generate speculative chemistry from whole cloth. Formal criteria gate what the agent surfaces; experts retain veto power.

Gambetta closes on outcomes:

"We believe that together with the DoE and its collaborators, we can unlock the scale of computing required to tackle the United States' most pressing scientific challenges."

What the public record establishes is a coherent thesis: Genesis wagers that agentic literature mining plus inverted algorithm-first matching can accelerate the path from proven quantum primitives to problems worth running on a Nighthawk queue.

### Sources

• IBM Research — IBM commits $50M in quantum access for US Genesis Mission (July 22, 2026)
• Quantum Computing Report — IBM Commits $50M in QPU Compute Access to Support U.S. DOE Genesis Mission (July 22, 2026)