There’s a Real Agentic AI Intelligence Community RFI. The Trend Coverage Missed It.
The agentic AI intelligence community story is usually told as a single trend piece. Three agencies are building agentic AI right now. The Army already ran a real, dated RFI asking for it. No coverage I found connects these two facts. If you are a BD lead or capture lead at an AI product firm, a systems integrator, or a DIB sub building agent-orchestration tools, a real Army buyer has already published a specific, benchmarkable requirements list, even though the window to respond has closed and no follow-on has posted yet.
The Defense Intelligence Agency, the National Geospatial-Intelligence Agency, and the FBI each have a Chief AI Officer on record describing agentic AI programs already in motion. DIA’s is already live. Separately, in June 2026, the Army posted a request for information asking industry for exactly this kind of platform. It closed a month later. Nobody wrote about both at once.
That gap matters for you specifically. The trend coverage tells you agentic AI is coming to the intelligence community. It does not tell you the Army already asked industry what it would take to build one. The RFI itself was issued through a university-affiliated research corporation working with Army offices, not directly by a contracting office, so whether a follow-on runs as a standard federal solicitation, a consortium or Other Transaction Agreement through that same intermediary, or a subcontract under its existing agreement is not yet known. I traced both threads and worked out what a reader should actually watch for next.
What DIA, NGA, and FBI Are Actually Building
Each agency described a different piece of the same shift: from single AI tools to networks of AI agents working together.
- DIA’s Chief AI Officer, Maj. Gen. Robert Kinney, said DIA ran a 90-day sprint to build its first enterprise AI platform service. ChatDIA, DIA’s AI assistant, is already deployed on JWICS, the intelligence community’s top-secret network.
- NGA’s Chief AI Officer, Michelle Aten, described an agentic framework built around specific tasks that subject matter experts identify. NGA is also working across the intelligence community to avoid multiple agencies building the same agents twice.
- FBI’s Chief AI Officer, Katie Noyes, described a role-based approach. Different AI agents support different analyst roles, including counterterrorism and cyber analysis.
None of the three officials gave a dollar figure. None named a specific contract behind these programs. This is program-formation evidence, not an open competition you can bid into today.
The Army RFI Nobody Connected to This Trend
While DIA, NGA, and FBI were describing agentic AI in the abstract, the Army had already asked industry for it in specific terms.
- The RFI posted June 4, 2026, titled “Agentic AI Platforms for Army Intelligence.”
- It was issued by the University of Arizona Applied Research Corporation, working with Army G-2 (Military Intelligence) and the Army Research Laboratory.
- Responses were due July 3, 2026, at 5:00 p.m. Eastern. That window has closed.
- The notice explicitly welcomed “traditional/non-traditional businesses, government agencies, and academic/nonprofit entities.”
That fourth point is the one worth sitting with. A classified-network RFI that explicitly names non-traditional and academic entities, not just cleared incumbents, is a real, textual signal that the Army wants options beyond its existing vendor base for this specific capability. The eligibility language reported for the notice does not name a clearance or an existing Army Intelligence relationship as a precondition. I have not verified the RFI’s full text against SAM.gov directly. A follow-on solicitation could still set its own eligibility terms. That could include facility-clearance requirements a JWICS deployment would normally carry.
What the RFI Actually Requires
The Army wants agentic AI platforms that let intelligence analysts build, deploy, monitor, and govern AI agents. Those agents need to reason, retrieve information, and take action through tool and function calls. All of it has to run inside classified networks.
| Deployment environment | Priority | What it means for a vendor |
|---|---|---|
| JWICS | Top priority | Top-secret network; highest security and accreditation bar |
| SIPRNet (IL6) | Second priority | Secret-level classified network |
| IL5 | Third priority | Controlled unclassified/sensitive environment, lower accreditation bar |
The RFI also flagged a specific technical constraint commercial AI platforms are generally not built for: edge and disconnected, intermittent, or limited-bandwidth use. An analyst in the field cannot count on a live connection back to a cloud model.
- Agents must reason over available information, not just retrieve and summarize it.
- Agents must take action through tool and function calls, not just generate text.
- The platform must run without persistent connectivity, on the edge, not only in a data center.
If your platform cannot do all three inside a classified environment, this specific opportunity is not for you yet. If it can, you have a concrete requirements list. Three separate IC agencies are independently investing in this same capability category. Only DIA’s JWICS deployment corroborates a specific line item on that list so far.
None of this is a brand-new accreditation regime built just for agentic AI. JWICS is an existing intelligence community network with its own established authorization process. SIPRNet and the IL5/IL6 tiers are existing classification levels under the DoD’s own cloud security rules. Either way, any software placed on these networks already has to clear a standard federal authorization process, agentic or not.
What is new sits on top of that baseline. CISA, NSA, and five partner countries published joint guidance on April 30, 2026, specifically for agentic AI. It calls for incremental rollout, starting with low-risk tasks. It calls for tight limits on what tools an agent can touch. It calls for continuous monitoring and a named human accountable for each agent’s actions. A platform that already clears standard classified-network authorization still has to show it meets this added layer.
Where a Nontraditional Entrant Most Likely Fits First
A small business or nontraditional entrant with no prior classified-network track record is unlikely to win a JWICS-accredited deployment on its own, on a first attempt. That does not make this RFI irrelevant to you. It changes where you should aim.
- Subcontracting under an existing IC prime that already holds classified-network accreditation is a more realistic near-term entry point than a direct prime award. Accreditation itself takes time you likely don’t have before a follow-on solicitation drops.
- The edge and disconnected-operation requirement is a specific technical gap. Commercial AI platforms are generally built around a persistent connection to a cloud-hosted model, a design assumption the RFI’s own edge/DIL language explicitly works against. A firm that has already solved reliable AI performance without a persistent connection has a real differentiator here, not just a checkbox, though no source reviewed confirms how many competitors have actually solved it.
- Watch for an Other Transaction Agreement rather than a traditional FAR contract on any follow-on. Army research organizations commonly use OTAs for early AI prototyping work, and this RFI’s own university-affiliated issuer makes a consortium OTA a plausible path. OTAs typically carry a lower bar to entry than a full and open FAR competition. No source reviewed confirms which vehicle a follow-on to this specific RFI would actually use.
Is This a Live Opportunity, or Just Signal?
Right now, it is signal, not an open door. The RFI’s own response window closed over a month ago. No follow-on solicitation, sources-sought notice, or draft RFP referencing it has surfaced in sources I reviewed.
Here is what is actually confirmed, separated from what is not:
- Confirmed: a dated, scoped RFI existed and closed. No confirmed dollar value or follow-on vehicle yet.
- Confirmed: DIA has an operational agentic AI tool, ChatDIA, on JWICS. No disclosed contract or dollar figure behind it.
- Not confirmed: any dollar figure specific to agentic AI at DIA, NGA, FBI, or the Army.
- Not confirmed: whether NGA’s separate $708 million SEQUOIA contract or CACI’s $290 million Luno-A contract touch agentic AI specifically. Both are real, disclosed AI/ML-in-intelligence awards, but neither source ties them to this capability.
Do not read those two contract figures as agentic AI money. They are the broader scale of AI spending inside the intelligence community, not evidence this specific capability is funded. This is what AI in government actually looks like at this stage: real signals from multiple agencies, no single confirmed contract yet.
FAQ
What is agentic AI, and how is the intelligence community actually using it?
The agentic AI intelligence community shift means AI systems that can reason, retrieve information, and take action on their own, not just generate text in response to a prompt. DIA, NGA, and FBI are each building agent-based tools instead of single AI assistants. IC leadership describes this as a response to intelligence data volume outpacing what one model or one human analyst can process, though the sources reviewed for this piece attribute that reasoning to agency leadership’s own statements, not to independently documented cause and effect.
Is there a real government solicitation for agentic AI platforms right now, or is this still just a trend story?
There was a real one. The Army posted an RFI for agentic AI platforms for Army Intelligence on June 4, 2026, with responses due July 3, 2026. That window has closed, and no follow-on solicitation has surfaced yet, but the RFI itself was concrete: a named issuing entity (the University of Arizona Applied Research Corporation, with Army G-2 and the Army Research Laboratory), a specific response due date, and a specific technical scope.
Which intelligence agencies are building agentic AI capabilities, and how far along is each one?
DIA is furthest along. Its ChatDIA tool is already deployed on JWICS, its top-secret network. NGA is building a task-based agentic framework and coordinating with other IC agencies to avoid duplicate work. FBI is building a role-based approach with different agents supporting different analyst functions. None has disclosed a specific contract or dollar figure for this work.
Can a non-traditional or commercial AI vendor actually compete for this work?
The Army’s RFI explicitly said yes. It welcomed “traditional/non-traditional businesses, government agencies, and academic/nonprofit entities,” language that does not assume an existing defense relationship or clearance history.
What would a follow-on Army Intelligence agentic AI solicitation likely require?
Based on the RFI’s own text, a platform that can be accredited for JWICS first, then SIPRNet (IL6) and IL5. It would need to let analysts build and govern AI agents capable of reasoning, retrieving information, and taking action through tool and function calls, and it would need to work without a persistent network connection for analysts operating at the edge.
BD classification: MONITOR Confidence: Moderate Evidence that would upgrade this: A follow-on Army solicitation, sources-sought notice, or draft RFP referencing the June 2026 RFI or its stated requirements, or a specific agentic-AI contract or OTA award from DIA, NGA, or FBI. Evidence that would downgrade this: Confirmation that the Army folded the RFI’s described capability into an existing contract through a modification, with no new competition planned.
