The app the program office never shipped
A decade ago a Ukrainian officer and a handful of volunteer programmers built an artillery fire-control app and handed it to gun crews on cheap tablets. Fire missions that took minutes now took seconds. The official program for the same capability worked its way through committees and died there. By the time CSIS surveyed Ukrainian units, the volunteer software was running in 90 to 95 percent of artillery units. The government version never fielded.
What the volunteers built was a mission application. It took the unit's own data and turned it into fire decisions at the speed the fight demanded. It ran on hardware the crews already had. And the unit owned it. Chatham House found that Ukraine's decentralized approach produced capability faster than any acquisition process on either side of the war. The volunteers were faster because they built for the user standing next to them, not because they were smarter than the program office.
The layer between the platforms and the radios
Defense software sorted itself into two stacks over the last decade. Enterprise C2 went to the platforms. Lattice took a $20B Army contract. Maven grew from $103M to a program of record serving combatant commands. Below them sit the transports. TAK, radios, mesh gear, the plumbing that moves a position report from one node to another.
Between those two stacks is where a mission actually runs. A team builds its loadout before stepping off, watches its cameras, and tasks people over whatever link is still up. Somebody asks a question and needs the answer with the source attached. Afterward the team reviews what happened and proves it to command, legal, or a partner force. That is the mission application layer, and every unit that could not wait has been building its own piece of it. The artillery app was one. The strike-coordination bots running on messaging apps are another. The layer keeps getting built by hand because nobody shipped it as a product.
What it takes to ship the layer as a product
AI is a mission utility now, and a mission application has to treat it like one. Every model call in EdgeLance routes through a single inference gateway. It checks capacity and classification policy before the model runs. It stamps each answer with the model hash, the quantization, and the policy decision that routed it. Any open-weight model plugs in. Any hardware the team already carries runs it, across Windows, Android, macOS, iOS, and Linux.
Mission state cannot live on a server nobody can reach. EdgeLance keeps the mission in a signed ledger that replicates peer to peer over the mesh. Every node holds a copy. Merges converge on their own. Lose a node and the mission keeps going, because there is no authoritative cloud to cut.
The mission also needs edges. Organization, workspace, and mission scopes bind every task, asset, and marker to the right context, with control leases to hand assets between operators cleanly. When the mission ends, the burn destroys the key and everything under it. The proof chain survives. And the compliance package builds itself. SBOM, control matrix, provenance attestation, generated from the audit ledger with every release. The paperwork that stalls most defense software comes off the same build that produces the software.
The layer already exists. We counted.
This week we counted the EdgeLance codebase at 629,791 lines. Over a thousand API endpoints. Twelve workspace modules covering the mission from plan to burn. The load-bearing pieces, the edge intelligence control plane, the burn ledger, and the mission pack compiler among them, are patent pending with the USPTO.
It packages for air-gapped delivery through Zarf and UDS, so it installs on a disconnected network from a single signed artifact. It syncs upward to Lattice, TAK, and partner systems when the team decides, and runs alone when the team decides that instead. The enterprise platforms consume what the edge produces. Owning the producing layer is the position.
Bottom-up layers win
Ukraine's lesson was never about tablets. Users adopt software that serves them and route around software that serves someone else. The volunteer app hit 90-plus percent adoption with no mandate, no fielding plan, and no budget line, because gun crews wanted it. The mandated alternative had all three and died anyway.
Enterprise platforms serve the command. The mission application layer serves the team in the fight, and syncs upward when the team chooses. Operators have been asking for this layer for a decade by building it themselves. Now it exists as a product, and it is called EdgeLance.