MOSAIC warfare is an architecture, not a weapon system
DARPA's MOSAIC concept replaces exquisite monolithic platforms with modular functional nodes that can be mixed and matched via AI-enhanced networks. Instead of an $80M aircraft doing ISR, EW, and strike, the idea is a network of cheaper nodes each contributing one or two capabilities, composed into mission plans in real time based on what is available. If one node is destroyed, the system recomposes around the gap. The force becomes a mosaic instead of a set of irreplaceable tiles.
The concept has generated real programmatic momentum. DARPA's 2026 RFI (DARPA-SN-26-33) requests autonomous drone warfare networks with large-scale unmanned constellations and containerized support systems. The technical requirements call for autonomous mission replanning, dynamic task allocation, formation reshaping, path optimization, edge-based computing, and collaborative multi-agent operations. Peraton Labs is developing self-healing networks for mosaic warfare. The XRQ-73 SHEPARD hybrid-electric reconnaissance drone flew at Edwards in April 2026.
The missing layer is between the node and the network
MOSAIC describes how nodes compose at the network level. What it does not specify is what runs on each node. The node needs local AI inference so it can contribute ISR capability without reachback. It needs mesh connectivity so it can participate in the network. It needs device management so it can be trusted. It needs evidence capture so its sensor data has provenance. It needs fleet management so its software stays current.
No current defense AI platform fills this layer. Enterprise C2 systems like Lattice and Maven operate above it, assuming network infrastructure and dedicated compute. Individual AI tools like object detectors or language models operate below it, providing single capabilities without the integration layer that ties them into a mission workflow.
The node-level operating layer, the software that turns a MacBook or a Jetson or a phone into a managed, AI-capable, mesh-connected tactical node, is the gap. That is the layer EdgeLance occupies.
What the DARPA RFI actually requires at the node
Read the technical requirements in DARPA-SN-26-33 and translate them to node-level software: 'edge-based computing' means local inference with no cloud dependency. 'Autonomous mission replanning' means the node has to make decisions when it cannot reach the network. 'Collaborative multi-agent operations' means nodes need to share data efficiently across unreliable links. 'Dynamic task allocation' means the mission can reassign nodes, which requires fleet management that can push new model loadouts and configuration in the field.
These are not enterprise platform requirements. They are node-level requirements. The MOSAIC concept works when each tile in the mosaic is independently capable and can be composed with other tiles through a shared mesh. EdgeLance provides that capability: local AI, mesh routing, device management, evidence capture, and fleet control on each node.
Composability requires interoperability
A MOSAIC node has to feed data upstream to whatever command system is in use. EdgeLance publishes entities via Cursor-on-Target for TAK/ATAK interoperability. It syncs structured events, evidence, and audit trails through the mesh when connectivity allows. It produces a mission record that larger systems can consume.
The node does not need to replace the enterprise layer. It needs to be a useful, managed, auditable participant in a larger system. That is a different design goal than building another C2 platform, and it is the design goal the MOSAIC concept actually demands.
The DoD has spent billions on the network layer and the enterprise layer of MOSAIC. The node layer, the software that makes each tile in the mosaic capable and trustworthy, is where the gap is. It is also where EdgeLance has been building from the start.