Fix Navy R&D: AI-Speed Reform at the Office of Naval Research

A decade is an unacceptable delivery schedule for most technologies—especially the ones meant to deter conflict and win fights. Yet that’s the timeline defense stakeholders routinely cite when they talk about how long it can take to move a promising idea from a lab to a fielded capability.

That’s why the Navy’s new leadership at the Office of Naval Research (ONR) is getting attention well beyond the usual Washington personnel churn. Observers say acting chief Rachel Riley brings an unusual mix: an appetite for organizational reform shaped by time in a Department of Government Efficiency (DOGE)-style environment, plus private-sector operating rigor from years of consulting. Whether you like DOGE or hate it, the underlying problem Riley inherits is real: Navy research has too many programs that start strong, drift, and never land in fleet-relevant outcomes.

This post is part of our AI in Defense & National Security series, and I’ll take a clear stance: the fastest path to meaningful R&D reform isn’t “more AI projects.” It’s using AI to change the operating system of defense research—what gets funded, how progress is measured, and when programs get stopped.

Why ONR reform is suddenly urgent (and measurable)

ONR reform is urgent because the Navy is competing against adversaries who iterate quickly, adopt commercial technology aggressively, and aren’t burdened by the same approval layers. The result is a widening speed gap—not just in autonomy and AI, but in how fast organizations can learn.

The measurable symptom is simple: cycle time.

When R&D timelines stretch into the 8–12 year range, three things happen:

• Threat models change faster than requirements documents. A system designed for yesterday’s operating environment arrives into today’s countermeasures.
• Vendors optimize for contract survival, not deployment. If “continued study” pays, “shipping capability” becomes optional.
• Program risk hides in plain sight. When milestones are vague, leadership can’t distinguish a tough technical challenge from a stalled effort.

ONR sits at an awkward intersection: it funds science that truly needs patient capital (oceanography, undersea sensing, certain cryptographic work), while also sponsoring efforts that look like a slower, more expensive version of what commercial teams already build.

Riley’s critics point to DOGE’s disruptive reputation and previous efforts to cut large numbers of staff in other agencies. Her supporters point to a different claim: strong leadership is required to say “no,” stop pet projects, and demand timelines that match operational reality.

The real problem isn’t research—it’s the “handoff”

The bottleneck isn’t that smart people can’t invent. The bottleneck is the handoff from research to adoption.

From prototype to program of record: the graveyard in the middle

Defense R&D has an infamous middle zone:

• Promising demos happen in controlled environments.
• Requirements harden.
• Testing expands.
• Budgets shift.
• A new stakeholder adds a new condition.

Then the effort becomes a permanent science project.

A useful way to describe this is: the Navy doesn’t just need better ideas; it needs better throughput. Throughput is an organizational design problem.

What “fleet relevance” should mean in 2026

Fleet relevance isn’t a PowerPoint phrase. It should translate into three concrete questions ONR leadership can enforce:

1. Who is the operational owner? If no commander would sign for it, you don’t have a customer.
2. What’s the adoption path? Transition partner, contracting route, sustainment concept.
3. What’s the timebox? A calendar date that forces trade-offs.

This is where AI can help—not as an “AI widget,” but as a management tool that makes drift harder.

How AI can accelerate Navy modernization without creating more noise

AI accelerates modernization when it shortens decision loops and reduces uncertainty. It fails when it becomes another category of pilot programs that never transition.

Here are four practical AI-enabled reforms that fit ONR’s role and the Navy’s realities.

1) Use AI for portfolio triage: “What should we stop?”

The hardest reform is stopping work that has supporters.

A modern ONR portfolio can include hundreds of efforts across basic research, applied research, and advanced technology development. Leaders need a repeatable way to answer:

• Which projects are duplicative with commercial capability?
• Which are blocked by integration (not science)?
• Which are scientifically promising but operationally irrelevant?

AI-assisted portfolio triage can ingest program artifacts (reports, milestone histories, test results, spend profiles, dependency maps) and produce risk signals:

• schedule slip probability
• dependency bottlenecks
• vendor concentration risk
• “transition readiness” score based on objective evidence

Used correctly, these tools don’t replace judgment—they give leaders the confidence to act earlier.

A portfolio that never cancels projects isn’t “supporting innovation.” It’s refusing to choose.

2) Make “commercial first” real with machine-speed market intelligence

One of the sharp critiques in the source article is that ONR sometimes under-asks: What’s commercially available?

A credible “commercial-first” posture requires constant scanning:

• patent and publication trends
• venture funding patterns
• supplier chain maturity
• export controls and IP constraints

AI can automate much of this. Think of it as defense market intelligence built for program managers: not generic trend reports, but product-level comparisons that answer, “Could we buy this, integrate it, and harden it faster than we can invent it?”

This matters most in areas like:

• autonomy stacks and perception
• simulation and synthetic data
• secure networking and edge compute
• counter-UAS sensing and fusion

3) Shrink test-and-evaluation timelines with digital engineering + AI

If you want shorter timelines, you have to attack test throughput.

AI-enabled test modernization typically looks like:

• digital twins for ships, unmanned platforms, and sensor systems
• automated log analysis to find anomalies in hours, not weeks
• model-based systems engineering that keeps requirements traceable
• synthetic environments to stress autonomy and cyber behaviors

This is especially relevant to the Navy because maritime operations involve complex environments—weather, sea state, cluttered littorals—where collecting “real data only” can be slow and expensive.

The goal isn’t to avoid real-world testing. It’s to arrive at real-world testing with fewer unknowns.

4) Build trustworthy AI governance into R&D from day one

Defense AI that can’t be trusted won’t be adopted. Period.

If ONR is going to fund AI for mission planning, intelligence analysis, undersea autonomy, or cyber defense, governance can’t be stapled on later. Strong programs define early:

• data provenance (what data, from where, under what rights)
• evaluation standards (robustness, drift, adversarial resistance)
• model deployment constraints (edge vs cloud, latency, offline modes)
• human-in-the-loop decisions (what must remain human authority)

The Navy’s environments—disconnected, contested, safety-critical—make this non-negotiable.

Where ONR must not mimic Silicon Valley

Speed is not the only metric that matters.

ONR funds research that commercial investors won’t touch because the payoffs are too distant, too niche, or too regulated. Undersea physics, ocean climate science, marine geosciences, specialized materials, and certain encryption research are all examples of work that can be strategically decisive but commercially unattractive.

Here’s the balancing act Riley (and any ONR leader) has to get right:

Keep the “strategic science” lane protected

Strategic science needs:

• stable funding
• long horizons
• tolerance for failure

Trying to force everything into 18-month product cycles will kill the very work that preserves U.S. advantage.

Demand delivery in the “applied capability” lane

Applied programs should have:

• clear transition partners
• timeboxed milestones
• integration plans

Different lanes. Different rules.

If you’re running ONR like a single monolith, you’ll either:

• slow down applied delivery, or
• starve basic science

Both outcomes are bad for national security.

A practical playbook for leaders: 90 days to change the trajectory

If you’re a defense R&D leader, a program executive, or a contractor supporting Navy modernization, here are moves that produce real change fast.

1) Publish a cancellation threshold

Define what triggers a stop or restructure. Examples:

• two consecutive missed milestones without validated root cause
• no transition sponsor by a fixed date
• inability to demonstrate progress against an operational metric

This reduces the politics. Everyone knows the rules.

2) Require a “commercial equivalence memo” for relevant projects

Before funding applied work in autonomy, analytics, comms, or simulation, require a short memo:

• best commercial alternatives
• what’s missing for Navy use
• why ONR-funded R&D is the fastest path

It’s amazing how much waste disappears when teams must answer this in writing.

3) Standardize AI evaluation artifacts

For AI-related efforts, enforce a minimal set:

• dataset description and access rights
• test harness and metrics
• adversarial and robustness tests
• deployment constraints

This prevents “demo theater.”

4) Create an ONR-to-Fleet transition scorecard

Track projects by:

• time to first fleet experiment
• time to operationally relevant environment
• time to program adoption decision

Reward teams that move capability, not paperwork.

What this leadership moment means for AI in Defense & National Security

Leadership changes don’t modernize a military on their own. What matters is whether leadership uses the moment to install mechanisms that persist after the headlines fade.

Riley’s arrival (and the reactions to it) highlights a deeper truth: AI advantage in national security depends as much on acquisition and R&D throughput as it does on model performance. A Navy that can’t adopt faster than adversaries can adapt will always be late—no matter how brilliant its research teams are.

If your organization builds, buys, or integrates AI for defense—mission planning tools, autonomy stacks, intelligence analysis platforms, cyber defense analytics—this is the right time to align with ONR’s likely direction:

• clearer outcomes
• tighter timelines
• more commercial awareness
• higher expectations for measurable progress

The question worth sitting with is simple: If ONR demanded proof of transition readiness every quarter, would your AI program look stronger—or would it collapse under the light?

---

If you’re working on AI for defense modernization and want a concrete transition plan—from data strategy to evaluation to deployment constraints—I can help you map the shortest credible path from prototype to operational use.