PSP Architecture
The DoW AI Strategy's 7 Pace-Setting Projects form an interconnected architecture for defense AI transformation. Based on the Kilger memo (January 2026).
DoW AI Strategy — Kilger Framework
The Department of War AI Strategy identifies 7 Pace-Setting Projects (PSPs) that collectively build the AI infrastructure for autonomous warfare. These are not independent programs — they form a layered architecture where warfighting PSPs (Swarm Forge, Agent Network, Ender's Foundry) depend on infrastructure PSPs (Open Arsenal, Project Grant) which in turn depend on enterprise foundations (GenAI.mil, Enterprise Agents).
Total PSPs
7
5 initiated/scalingTotal Investment
$38.3B
across FYDPStrong Dependencies
7
of 14 total connectionsHigh-Risk PSPs
2
Swarm Forge, Agent NetworkPSP Interconnection Architecture
Click a node to see details. Solid lines = strong dependencies. Dashed lines = weak dependencies. Arrows show data flow direction.
PSP Decision Points
Observable tripwires for each Pace-Setting Project
| PSP | Tripwire | Year | If Met | If Missed | Status |
|---|---|---|---|---|---|
| Swarm Forge | First 10K swarm-capable drones fielded | 2027 | Swarm CONOPS validated; production pipeline working | Swarm coordination technology not mature; falls back to individual drone ops | on-track |
| Swarm Forge | Multi-domain swarm exercise | 2028 | Multi-domain swarming is real; force design implications confirmed | Domain-specific swarms only; integration gap persists | on-track |
| Agent Net | Kill chain compressed to <15 min in exercise | 2027 | Decision superiority achievable at operational level | AI decision support adds latency rather than compressing it | on-track |
| Agent Net | Deployed to operational COCOM | 2028 | AI decision support operational; trust established | Commanders don't trust AI recommendations in real operations | on-track |
| Ender's Fndry | AI adversary passes Turing test in wargame | 2027 | Wargaming revolution; continuous AI-driven force design optimization | AI adversaries remain predictable; human wargaming still required | at-risk |
| Open Arsenal | AI module integrated to legacy platform in <90 days | 2027 | Modular AI architecture works; spiral upgrades viable | Integration still takes years; vendor lock-in persists | on-track |
| Proj Grant | >85% MCR on AI-maintained platforms | 2029 | Predictive maintenance transforms readiness; O&S savings validated | Sensor-to-prediction pipeline slower than modeled; data requirements underestimated | on-track |
| Proj Grant | 15% O&S cost reduction validated | 2030 | Business case for fleet-wide deployment proven | O&S savings less than projected; implementation costs offset gains | on-track |
| GenAI.mil | 50% of new DoW SW with AI assistance | 2028 | Software revolution dominant in defense acquisition | AI code gen remains niche; validation bottleneck constrains adoption | on-track |
| GenAI.mil | Secure IL6 GenAI platform operational | 2027 | GenAI accessible at classified levels; mission-critical applications possible | Security concerns prevent classified deployment; GenAI limited to unclassified | on-track |
| Ent Agents | 40% admin workflow automation | 2029 | 50K+ personnel freed for operational roles | Bureaucratic resistance and system integration challenges slow adoption | at-risk |
Critical Path Analysis
Foundation Layer (Must Succeed First)
- GenAI.mil (PSP-6): Provides the AI infrastructure backbone. Agent Network and Enterprise Agents both depend on GenAI.mil's secure LLM platform. If GenAI.mil fails, 3 other PSPs are compromised.
- Open Arsenal (PSP-4): Creates the integration standard. Without open architecture, Swarm Forge platforms can't receive AI module upgrades and Project Grant sensors can't plug into legacy platforms.
Warfighting Layer (Delivers Combat Value)
- Swarm Forge (PSP-1): The highest-cost, highest-risk PSP. Depends on Open Arsenal for modularity and Agent Network for C2. Delivers the expendable mass layer of the force pyramid.
- Agent Network (PSP-2): The “nervous system” of the autonomous force. Compresses kill chains and enables multi-domain coordination. Single point of failure for the entire architecture.
Key Risk: Simultaneous Execution of 7 PSPs
Historical evidence suggests DoW struggles to execute more than 2-3 major transformation programs simultaneously. The PSP architecture requires 7 to succeed in parallel, with strong interdependencies. Partial success (4-5 PSPs delivering) may be sufficient for meaningful capability improvement, but the full architecture effect requires all 7. The critical risk is not any single PSP failing but cascade failure through the dependency chain: if GenAI.mil or Open Arsenal underdeliver, warfighting PSPs are degraded. Congressional budget support across multiple FYDP cycles is the binding political constraint.
Source: Kilger memo analysis | Data as of Feb 2026
Methodology
PSP architecture analysis based on the DoW AI Strategy (Kilger memo, January 2026). Cost estimates are analyst-derived from program analogies, service budget justification documents, and FYDP projections. Dependency mapping based on technical architecture analysis and historical program interdependency patterns. Tripwire assessments use the same confidence framework as the broader study (high/medium/low). All costs in FY2024 constant dollars.
Tim Booher | Feb 2026