Emergency management is often described as slow to adopt technology, but the record doesn't really support that:
- GIS moved through the field quickly
- Drones became a part of damage assessment
- Starlink was running in EOCs soon after it was available
What the field has been reluctant to rely on is technology that fails when the disaster arrives. Any product that depends on stable connectivity is unavailable at precisely the moment it is needed. This is a design requirement, not a cultural obstacle, and practitioner research in the new AIDE Report from the Markle Foundation, Aspen Digital, and RAND makes the same point: many of these tools assume connectivity that a disaster removes.
The Market Is Aimed at the Wrong Phase
The AIDE Report also includes a finding that should give pause to anyone building in this space. RAND identified more than 1,100 AI-enabled products from more than 700 companies with some relevance to emergency management. The large majority are aimed at response (hazard monitoring, damage assessment, situational awareness), while comparatively few address preparedness, recovery, or cross-agency coordination.
The supply side has concentrated on the phase where adoption is most difficult, and has largely overlooked the phase that determines whether adoption happens at all.
Adoption Happens in Blue Sky
That matters because tools are not adopted during an activation. Nobody learns new software at landfall. Anything used in gray sky earned its place in blue sky: in the plan, in the training curriculum, in the exercise. If a tool is not part of an ordinary Tuesday, it will not be reached for at two in the morning on day three.
The AIDE research states the underlying problem directly; useful tools exist and emergency managers can identify valuable uses for them, but many agencies do not yet have the capacity or the trust to use them responsibly. Both capacity and trust are built during preparedness.
Budget Structure Is a Barrier Before Skepticism Is
Funding follows the same pattern. Aspen Digital's Jeremy Greenberg has noted that many AI vendors price by tokens, a model that does not align with how governments purchase. The surrounding conditions compound it: EMPG carries a 50% cost-share requirement, grant funding tends to favor equipment over recurring subscriptions, and FEMA's grant systems were unavailable for roughly five weeks during this year's funding lapse.
A one- or two-person county office has limited ability to commit to a recurring per-seat contract under those constraints. Adoption is in large part a question of budget structure rather than willingness.
Trust Has a Recent History
Trust has its own recent history, and the technology industry contributed to it. For many emergency managers, the first significant encounter with generative AI was not a product but a problem: fabricated disaster imagery circulating faster than official information during the 2024 hurricanes, false claims about federal assistance, and threats directed at personnel in the field.
Add that EOC records are subject to public disclosure and post-incident litigation, and that responsibility for an evacuation decision rests with the emergency manager rather than the vendor. Caution in that context is reasonable.
Start Where an Error Is Recoverable
All of which suggests an unglamorous starting point. The most practical early applications are the ones where an error is recoverable: after-action reports, Public Assistance documentation, hazard mitigation plan updates, exercise materials, grant applications, and public information in multiple languages.
This is repetitive, deadline-driven work that consumes a real share of a small office's week, and it passes through human review before it goes anywhere. Demonstrated value there is what builds the case for being trusted closer to consequential decisions.
Where This Leaves Both Sides
For vendors, that means building for preparedness, pricing in a way governments can actually buy, and designing for degraded connectivity. For emergency managers, the practical test is whether a tool holds up in an exercise.
The AIDE Initiative's summit in Austin next month is a reasonable place for both conversations to continue.