Thunderforge

Thunderforge is the US Department of Defense’s attempt to bring AI agents into the slow, document-heavy work of military planning. Announced on 5 March 2025 by the Defense Innovation Unit, the programme builds software that helps commanders develop campaign plans, allocate resources across a theatre, run wargames, and generate and refine courses of action — work that today still leans on decades-old tools and manual processes. The pitch, in the words of DIU’s programme lead, is to close “a fundamental mismatch between the speed of modern warfare and our ability to respond.”

The programme is built by a trio of companies. Scale AI is the prime contractor, supplying the agentic applications and the test-and-evaluation expertise to vet generative models for military use. Anduril contributes its Lattice software, into which Scale’s models are integrated for mission planning and simulation. Microsoft provides the underlying large language models. The contract’s value was not disclosed.

Thunderforge is being fielded first at US Indo-Pacific Command and US European Command — the two theatres where planning against a large, capable adversary is most demanding — with the intent to scale to other combatant commands if it proves itself. It sits alongside the Pentagon’s wider push to weave generative AI into command and control, and it is one of the clearer signs that the technology’s military use is moving from back-office experiments toward the heart of how the armed forces plan to fight. As with much of this work, the open questions are less about whether the agents can produce plans than about how far human planners will come to trust them.