Fix SBIR: AI-Powered Paths to Field Defense Tech

A 2024 Government Accountability Office report counted 1,100+ open recommendations for the Department of Defense—a blunt indicator of how hard it is to change a massive enterprise. If you work anywhere near defense R&D, you don’t need another report to tell you the same story: promising tech gets funded, demoed, praised… and then it dies quietly before it reaches operators.

That failure pattern has a name: the valley of death, the gap between research dollars and real procurement. The War on the Rocks argument for a “clean sheet” redesign of the Small Business Innovation Research (SBIR) program lands on a simple point I strongly agree with: patching SBIR won’t fix SBIR. The incentives are wrong, the handoffs are weak, and transition is treated like an optional extra.

This post sits in our AI in Government & Public Sector series, where we look at practical ways AI improves public outcomes. Here, the practical goal is straightforward: get useful defense technology from small businesses into programs of record faster—without turning “innovation” into theater.

The real SBIR problem: transition is nobody’s job

Answer first: SBIR struggles because it funds exploration well but makes transition someone else’s problem—usually a program office that’s measured on cost, schedule, and performance, not tech scouting.

SBIR was built to stimulate innovation and commercialization. In defense, it also functions as one of the few real entry points for non-traditional vendors. That part matters. The reality, though, is that Phase I/II awards often produce prototypes without a funded runway to become capabilities. Program managers (PMs) aren’t villains; they’re rational actors responding to incentives. If your job is to keep a major program on schedule, adopting an external prototype can look like volunteering for risk.

The RSS source calls out several root causes that show up repeatedly across defense modernization efforts:

• PM disengagement: SBIR projects run adjacent to programs of record instead of inside their execution logic.
• A funding chasm: R&D money exists; procurement money exists; the “in-between” is weak.
• Under-supported technical oversight: Technical Points of Contact can be overtasked and undertrained in program management.
• Innovation offices competing for turf: Multiple “front doors” dilute accountability and confuse vendors.
• Bridge funds that don’t connect to a receiver: Stopgap pools can look helpful while masking the absence of a real transition owner.

If you’ve been around defense acquisition long enough, you’ve seen a prototype celebrated at a demo day and then orphaned because no one had budget authority—or political cover—to adopt it.

What should stay: Phase III is the sharpest tool in the kit

Answer first: Keep SBIR’s Phase III sole-source authority and make it easier to use.

The most fieldable part of SBIR is also the most misunderstood: Phase III. It’s a flexible pathway to follow-on contracting for capabilities that emerged from SBIR work. When it’s used well, it compresses timelines and reduces paperwork friction.

But in many organizations, Phase III is treated like an exception instead of a default transition mechanism. That’s less about law and more about muscle memory, templates, documentation, and confidence.

The clean-sheet proposal that actually changes incentives

Answer first: Put transition money inside programs of record and force usage through accountability—then SBIR becomes a feeder system, not a science fair.

The RSS article’s core design move is the one that matters most: create a dedicated, flexible transition funding line inside each acquisition program, sized at 10% of that program’s R&D budget, usable specifically for transitioning new technologies.

That does two things at once:

1. Creates a real “receiving dock” for SBIR outputs (and other emergent tech).
2. Makes program offices the owners of adoption, not spectators.

I like this approach because it aligns authority and responsibility. If the PM owns cost/schedule/performance, then the PM also needs a budget mechanism that makes tech insertion normal, not heroic.

Why “colorless” transition funds beat more pilot programs

Answer first: Transition fails when money is tied to the wrong side of the gap; “colorless” program-side funds reduce the bureaucratic contortions that stall adoption.

Defense organizations often respond to transition failures by creating a new office, a new fund, or a new sprint process. That can create motion without throughput.

A program-side transition line is different. It’s not another demo. It’s not another grant. It’s budgeted intent.

This is also where AI fits naturally: if you fund adoption inside the program, then AI-enabled tools can optimize how candidates are found, evaluated, and contracted—without pretending AI can fix incentive misalignment on its own.

Where AI actually helps (and where it doesn’t)

Answer first: AI won’t “solve acquisition,” but it can remove three major bottlenecks: topic quality, portfolio alignment, and contracting paperwork.

AI is most useful in defense innovation when it targets repeatable administrative and decision workflows—especially the ones that slow down humans who already know what they want.

Here are high-impact applications that map directly to the redesign ideas in the RSS content.

1) AI-assisted topic development that’s plain-language and operator-led

Answer first: Use AI to turn messy operational input into clear problem statements, then let humans vote and refine.

Today’s topic processes often reward insiders: a narrowly worded topic can unintentionally (or intentionally) favor a vendor already in the loop.

A better model is:

1. Operators submit pain points in plain language.
2. AI clusters similar needs across units/commands.
3. Humans (operators + program staff) refine and vote.
4. Program offices participate in final selection to ensure a transition landing zone.

That combination is powerful because it improves fairness and increases adoption probability. The goal isn’t “more topics.” It’s fewer, clearer problems with a real customer on the back end.

2) AI-driven portfolio alignment to programs of record

Answer first: Every SBIR award should map to an existing or planned program and a transition plan; AI can enforce that mapping at scale.

Portfolio management is unglamorous, but it’s where most innovation ecosystems fail. If you can’t answer “who will buy this?” you’re funding hope.

AI can help by automatically:

• Tagging proposals to relevant programs, platforms, and mission areas
• Flagging duplicates (“We’ve funded this exact idea three times”)
• Identifying missing enablers (test environment, cyber approvals, data rights)
• Predicting transition risk based on historical patterns (e.g., long approval lead times)

Humans still decide. But AI can stop the portfolio from drifting into disconnected experiments.

3) A Phase III marketplace that makes contracting fast and compliant

Answer first: A searchable Phase III marketplace plus contract-generation automation is the quickest way to turn “interest” into an award.

The RSS article calls for a centralized, AI-enabled Phase III marketplace where program managers can discover SBIR work and generate compliant awards quickly.

That’s exactly right. The friction isn’t just choosing a vendor—it’s generating the documentation trail, assembling past performance artifacts, aligning terms, and getting the right signatures.

A marketplace model should include:

• Standardized capability summaries written for acquisition teams (not engineers)
• Evidence packages: test results, security posture, integration notes
• Data rights posture in plain English
• Templates for Phase III justification and award packages

If a PM can move from “I want that” to “award in progress” in days instead of quarters, you’ll see adoption behavior change.

What to do with “innovation offices” and bridge funds

Answer first: Consolidate overlapping innovation entities and judge them by barriers removed and transitions achieved—not by the number of events, topics, or pilots.

The RSS content argues for merging fragmented organizations and eliminating slush/bridge funds that sit on the wrong side of the valley. I’m sympathetic.

Fragmentation creates three predictable problems:

• Vendors waste time navigating multiple front doors
• Offices compete for credit instead of coordinating transition
• Money gets spread thin across pilots with no receiving program

A cleaner operating model is:

• One transition institution with a clear mandate: invest + transition
• Program offices own adoption via the program-side 10% transition line
• Innovation offices function as smoothers: security approvals, test access, integration pathways, contracting help

A useful metric for innovation offices: “How many weeks did you remove from time-to-field?”

That’s measurable. And it keeps everyone honest.

A practical redesign checklist leaders can use in 2026 planning

Answer first: If you’re serious about SBIR reform, implement governance that makes transition the default and non-transition the exception.

Whether or not the full clean-sheet redesign happens, you can adopt its most effective elements during annual planning, POM builds, and portfolio reviews. Here’s a concrete checklist I’ve found works for leaders who want results without waiting for perfect legislation.

Governance rules that create real throughput

1. Tie every topic to a program of record (or named future program) with a written transition plan.
2. Create a program-side transition budget line (the 10% concept) and require quarterly execution reporting.
3. Require a transition rate threshold for repeat performers and for government technical managers (the RSS article suggests 10%).
4. Cap perpetual Phase I/II farming (the article proposes $25M per company/subsidiaries) to prevent dependency.
5. Mandate performance reporting so small businesses leave with credible past performance artifacts.

The AI enablement layer (what to procure or build)

• Topic clustering and plain-language rewriting with human review
• Portfolio “fit scoring” against programs of record and mission needs
• Automated document assembly for Phase III award packages
• Marketplace search that matches needs to proven SBIR outputs

You don’t need to “AI everything.” Pick the two workflows that burn the most staff time and start there.

Why this matters beyond SBIR: AI adoption is a national security workflow problem

Answer first: The U.S. doesn’t mainly lose time on invention; it loses time on adoption, integration, and scaling—and that’s where AI-enabled process redesign pays off.

In the AI in Government & Public Sector conversation, it’s tempting to focus on models, autonomy, and algorithms. Defense modernization lives or dies somewhere less glamorous: budgets, approvals, contracting, integration, and sustainment planning.

That’s why the SBIR debate is bigger than SBIR. It’s a test case for whether the Department of Defense can turn promising technology—AI included—into fielded capability on timelines that match the threat.

If you’re leading an innovation shop, a program office, or a defense tech company, the question to ask in 2026 isn’t “How do we fund more prototypes?”

It’s this: What’s our repeatable path from operational need → award → integration → fielding—and which parts should be automated?