White House Science Chief Lays Out a Plan to 10x

The official target is 2x. The private aspiration, Kratsios admits, is probably 10x. That gap between what a government can say publicly and what its own architects believe is possible tells you most of what you need to know about where the White House Office of Science and Technology Policy stands on AI-accelerated research — and how hard the politics of ambition actually are.

Michael Kratsios, the 13th director of OSTP and science adviser to President Trump, sat down with Peter Diamandis on the Moonshots podcast to walk through three interlocking initiatives he's been building: the America AI Action Plan, the Genesis Mission, and a blueprint called "Science and a New Golden Age." Taken together, they represent the administration's most detailed public argument that scientific productivity is a policy variable, not a fixed feature of how knowledge gets made.

The conversation is worth engaging seriously, even for readers who will find its boosterism bracing. Kratsios is not a politician reading talking points. He thinks carefully about mechanism design, regulatory friction, and the gap between intent and outcome. Whether his proposals are sufficient to the problem is a different question — and one the interview mostly avoids.

The Diagnosis

Kratsios frames the problem as "Eroom's Law" — Moore's Law spelled backwards — the well-documented observation that drug discovery costs have roughly doubled every nine years, even as underlying tools have improved. His argument is that the scientific enterprise has calcified around a funding and publishing structure built for a different era.

"We just are not experimenting enough in the way that we conduct science," he told Diamandis. "We're doing the same thing over and over again and just putting more money towards it and believing that the outcomes are going to get better."

The structural critique is genuinely bipartisan, even if the proposed remedies are distinctly this administration's. The funding cycle forces researchers to begin applying for their next grant before finishing work on the current one. Peer review panels, Kratsios argues, systematically defund unconventional ideas because consensus-based committees trend toward the defensible. And the tenure pipeline means that early-career researchers — often those most likely to attempt the ambitious and fail productively — spend their most generative years navigating institutional gatekeeping rather than running experiments.

None of this is new. The critique of NIH grant culture has been a staple of science policy conversations for at least two decades. What's new is the claim that AI changes the calculus enough to make now the moment to act.

The Proposed Fixes

The Golden Age report, which Kratsios co-authored, proposes several mechanisms for disrupting grant-making orthodoxy. Three of them are worth examining closely.

Long-duration grants. The current default is roughly 18-month grant cycles, calibrated to institutional review timelines rather than the pace of discovery. The proposal: offer five-year grants, fully funded on day one, for research problems that require sustained exploration. The logic is straightforward — if you want scientists focused on the science, stop making them fund-raise continuously.

Fast grants. Small proof-of-concept awards, reviewed in under a month, for ideas that need rapid validation before they can attract larger funding. Kratsios frames this as regularizing a crisis response: rapid-decision grant-making showed what was possible under pressure, and there is no principled reason to reserve it for emergencies.

Golden tickets. This is the most structurally interesting proposal. Each member of a merit review panel would receive a small number of unilateral funding decisions — grants they can approve regardless of committee consensus. The intent is twofold: to create a pathway for genuinely unconventional ideas, and to attract stronger reviewers by giving the role real power. "Rather than having these people that are sort of like mid-tier people who just do this," Kratsios said, the golden ticket creates an incentive for elite scientists to participate in the review process.

The honest version of the critique here: golden tickets could just as easily concentrate idiosyncratic bias as unlock hidden genius. A unilateral funding mechanism is only as good as the judgment of the person wielding it. Kratsios acknowledges this implicitly when he talks about "meta-science" — the science of science — and the need to actually measure whether these interventions produce better research outcomes. Meta-science units have reportedly been announced at NIH and NSF. Whether they'll survive long enough to generate useful data is a separate question.

The Wilder Frontier

The most striking passage in the conversation is Diamandis reading back a section of the Golden Age report that envisions AI agents autonomously running scientific experiments, posting bounties for breakthroughs, verifying cryptographically signed results, and releasing payment via smart contract as milestones are met. A decentralized autonomous organization, essentially, for scientific discovery.

Kratsios didn't walk it back. "This opportunity gave this report gave us an opportunity to kind of dream big of where we could end up going," he said, "and I think it's something that is possible."

Possible is doing considerable work in that sentence. The vision presupposes that autonomous labs — where AI proposes hypotheses, robots run experiments, and the model designs the next iteration without human intervention — become sufficiently general to handle a wide range of experimental domains. Kratsios concedes the hardware isn't there yet, and that initial deployments will be narrow. Lila Sciences, which Diamandis mentions in the conversation, is building out significant autonomous lab capacity, but the distance between that and a machine-speed scientific marketplace coordinated by DAOs and prediction markets is not trivial.

What's interesting about including this vision in a government policy document is that it signals where the policy imagination is reaching, even if execution is years or decades away. Compare that to the EU AI Act, which Kratsios cites as a cautionary example of regulatory ambition outrunning technological reality — finalized before the models it was meant to govern even existed.

The 2x Problem

Here is where the conversation gets genuinely revealing. The Genesis Mission's official target is doubling scientific productivity over a decade. Diamandis pushes: is that just politically acceptable? Could it be 10x?

Kratsios is disarmingly candid. The 2x number, he explains, was set when the program was being designed in the prior year. By the time the Golden Age report was finalized, the team internally debated whether to revise upward — but had already committed publicly to 2x, so they let it stand. "My sense is," he said, "we should probably be aiming for 10."

That admission matters. It means the public target is lagging the administration's own internal assessment of what's achievable, and it suggests that the real constraint on ambition is not technology but institutional communication strategy. Government moves in 2x increments because 10x sounds unreliable. Whether that caution is wisdom or self-limitation is the central unresolved tension in everything Kratsios described.

The Geopolitical Frame

Woven through the entire conversation is a China-as-foil argument that by now has become structurally mandatory in any Washington tech policy discussion. The AI stack export program — turnkey American AI infrastructure for allied and developing nations, backed by Export-Import Bank and Development Finance Corporation financing — is framed explicitly as the answer to what Huawei was in telecom: a subsidized, geopolitically deployed technology layer that the U.S. had no comparable answer to.

The open-source dimension is the most interesting wrinkle. Kratsios acknowledges plainly that "the Chinese have very well-performing open source models" and that a cash-strapped American entrepreneur has a rational incentive to use them. The policy response — cultivate a domestic open-source ecosystem — is directionally sensible but offers no timeline and no mechanism beyond encouragement. The American AI Exports Program is further along; it has at least issued an RFP and is evaluating proposals. Whether that translates into a competitive alternative at global scale before Chinese open-source models become the default infrastructure for the Global South is an open question the interview doesn't try to answer.

What's Missing

The conversation is held between two people who share a fundamental commitment to technological optimism, which means certain questions don't get asked. There is no real engagement with who bears the downside risk when the "born free" regulatory posture proves to have been too permissive — when the assumption that a technology is safe to deploy freely turns out to be wrong. The EU AI Act's flaws are discussed at length; the costs of regulatory absence receive approximately zero scrutiny.

The labor question gets a similar soft treatment. Kratsios notes that the administration wants better data on AI's actual employment impact, and acknowledges — with unusual honesty — that companies may be labeling routine layoffs as AI-driven because "it plays better." That's a genuine insight. But the policy response remains at the level of data collection and voluntary upskilling programs. For a conversation that spent considerable time imagining AI agents running experiments at machine speed, the employment policy chapter is conspicuously incremental.

None of which makes the Genesis Mission unserious. The grant reform proposals are grounded in real structural problems. The autonomous lab vision, however speculative, is at least trying to imagine what AI-native scientific infrastructure actually looks like rather than just deploying AI onto existing processes.

Whether an administration that openly debates 2x versus 10x — and chooses 2x for communications reasons — can build the institutional momentum to achieve either remains the question that no amount of White House podcasting can answer in advance.

Samira Barnes covers technology policy and regulation for Buzzrag.