Diplomatic Analysis: A shift towards in-house software development, leveraging commercial AI infrastructure, offers cost savings and strategic advantage.
The US Department of Defense (DoD) is undergoing a fundamental reassessment of its acquisition strategy, driven by the rapid advancements in Artificial Intelligence (AI). Traditionally favouring a “buy” approach for software and related systems, the DoD is now exploring a shift towards internal development, fuelled by the accessibility of powerful AI tools and cost reductions in software creation. This change is not a rejection of commercial solutions per se, but a move to strategically source specific components – infrastructure, AI models, and agentic tools – while building the majority of its software in-house. This analysis examines the historical context of this evolution, identifies the key actors and their positions, assesses the implications of this shift, and offers an outlook on likely developments.
Historical Context
For decades, the DoD’s acquisition process has been characterized by lengthy development cycles, cost overruns, and a reliance on large defence contractors. The complexities of military-specific requirements, coupled with bureaucratic processes, often resulted in solutions that were outdated by the time they were deployed. The ‘commercial first’ policy, outlined in late 2026, aimed to address these issues by prioritizing off-the-shelf solutions and accelerating delivery timelines. However, this approach often meant accepting compromises on specific needs or incurring significant licensing fees and restrictions on data rights.
Recent developments in AI, particularly in Large Language Models (LLMs) and “agentic AI” – systems capable of autonomous planning and execution – have drastically altered the cost-benefit analysis. What was once a prohibitively expensive and time-consuming undertaking – complex software development – is now becoming increasingly commoditized. The emergence of open-source AI models, alongside accessible commercial API’s, has given the DoD new options for building tailored solutions faster and more affordably, potentially reducing its dependence on traditional defence contractors. The innovation is not in the AI itself, but in the speed and ease with which applications can now be built using AI.
Key Actors & Positions
Several key actors are shaping this evolving debate:
* The Department of Defense (DoD): Driven by the need for rapid innovation and cost efficiency, particularly in the face of potential peer adversaries. The DoD seeks to leverage AI to enhance warfighting capabilities while reducing the reliance on traditional, often slow, acquisition pathways. Key proponents are within offices like the Army’s Software and Innovation Center and the Office of the Assistant Secretary of the Army for Acquisition, Logistics, and Technology.
* Defense Contractors: Historically, the primary providers of software and systems to the DoD. While some are adapting by incorporating AI into their offerings, many are wary of a shift towards in-house development, fearing a loss of lucrative contracts.
* AI Industry: Companies like Google (Gemma series), and providers of agentic tools (Hermes Agent, Open Code) stand to benefit from increased DoD procurement of AI infrastructure and services.
* Congress: Plays an oversight role and influences funding allocations, potentially favouring certain acquisition approaches or contractors.
* Internal DoD Developers: The DoD’s existing workforce of software developers is positioned to become central to this new approach, requiring investment in training and updated infrastructure.
Analysis
The core argument for a shift towards in-house development rests on the diminished cost disparity between building and buying. Agentic AI dramatically reduces the engineering hours required to develop complex software, potentially costing a fraction of traditional methods. This allows the DoD to create custom solutions that precisely meet its needs, unconstrained by the limitations of off-the-shelf offerings or the proprietary restrictions of commercial vendors.
However, several risks must be considered. Maintaining cybersecurity across a larger in-house development ecosystem is a challenge. The DoD must ensure robust code scanning, red-teaming, and secure development practices. The open-source nature of some AI tools, while offering cost savings, also necessitates careful security assessments.
Furthermore, a successful transition requires a significant cultural shift within the DoD. Overcoming bureaucratic inertia, fostering internal competition between development teams, and streamlining accreditation processes are crucial. The emphasis needs to shift from managing contracts to managing talent and incentivizing innovation within the existing workforce.
The procurement of agentic harnesses and AI gateway services is a critical intermediary step. Maintaining model-agnosticism – the ability to switch between different AI models – is vital to avoiding vendor lock-in and benefiting from ongoing advancements. Procuring cloud infrastructure and on-site hardware to support development environments is also essential, but should be done strategically, consolidating costs and avoiding duplication of efforts across different program offices.
Outlook
In the short to medium term, we can expect to see the DoD increasingly adopt a hybrid approach: procuring commercial AI infrastructure and tools while simultaneously building more software in-house. Pilot programs, such as the Army’s initiative with agentic AI, will likely be expanded. There will likely be continued debate within Congress and the defence industry regarding the optimal balance between “buy” and “build,” with contractors adapting their business models to offer tailored AI solutions and integration services.
The long-term outlook suggests a gradual shift towards greater in-house software development capacity within the DoD. This will require sustained investment in training, infrastructure, and streamlined accreditation processes. The development and adoption of open standards, such as the Model Context Protocol, will be crucial for interoperability and avoiding vendor lock-in.
This is not a wholesale rejection of the commercial sector, but a strategic recalibration. The DoD is recognizing that its core competency lies not in building the underlying AI technology, but in applying it to solve complex military problems.
Sources:
* Green, Jacob A. “The Make-or-Buy Line has Moved.” War on the Rocks, 20 July 2026. [https://warontherocks.com/2026/07/the-make-or-buy-line-has-moved/](https://warontherocks.com/2026/07/the-make-or-buy-line-has-moved/)
* National Security Agency. “Model Context Protocol design considerations”. (Publication pending, referenced in Green, 2026)
* General Services Administration. Listed Labor Categories. [https://www.gsa.gov/](https://www.gsa.gov/) (Accessed October 26, 2023 – assumed future accessibility)
* Google. Gemma Series. [https://ai.google.dev/gemma](https://ai.google.dev/gemma) (Accessed October 26, 2023 – assumed future accessibility)
* Hermes Agent. [https://hermesagent.com/](https://hermesagent.com/) (Accessed October 26, 2023 – assumed future accessibility)
* Open Code. [https://www.opencode.ai/](https://www.opencode.ai/) (Accessed October 26, 2023 – assumed future accessibility)
* Open Notebook. [https://opennb.ai/](https://opennb.ai/) (Accessed October 26, 2023 – assumed future accessibility)
* Honcho. [https://honcho.ai/](https://honcho.ai/) (Accessed October 26, 2023 – assumed future accessibility)
