Speed, Flexibility, and the Price of Losing Leverage

Choosing an acquisition route for public-sector AI is not a contest between a “modern” option and a “bureaucratic” one. The relevant question is what each route makes easier, what it leaves uncertain, and whether the agency can restore the missing control before the capability reaches a consequential workflow. This chapter compares the routes identified in federal acquisition evidence—agency-directed and vendor-driven, custom and established, FAR-based and other agreements, service and product—using the criteria that matter after award: mission fit, speed, competition, transparency, data rights, testing, cost, and reversibility.

Four routes, one decision problem

An agency-directed acquisition begins with a mission need that the agency has defined and takes to the market. A vendor-driven acquisition begins when a supplier presents a capability that appears useful even though the agency did not formulate an AI-specific requirement at the outset. Neither origin is automatically responsible or irresponsible. The difference is where the initial problem definition comes from and how much time the agency has had to prepare for the technology.

GAO describes both patterns. The Department of Homeland Security used an established schedule to acquire generative AI capabilities for developing portions of hazard-mitigation plans, a case in which the agency had a defined public task. Elsewhere, the Department of Veterans Affairs awarded a task order for medical software with AI capabilities, and the General Services Administration acquired maintenance-tracking software whose vendor added a chatbot as an extra capability rather than in response to a defined requirement. In the latter pattern, the agency may receive useful functionality quickly, but the AI-specific questions can arrive after the procurement frame has already been set.

The advantage of an agency-directed approach is deliberate preparation. Officials can describe the outcome, bring in technical and mission experts, compare vendors against meaningful factors, and negotiate terms that fit the use. The cost is time. In a rapidly changing market, a carefully specified requirement can take long enough that the leading capability has moved, the vendor landscape has shifted, or the original business case no longer looks attractive. Agency direction is therefore valuable when mission specificity and public consequence justify the preparation, not because an agency-originated requirement is inherently superior.

A vendor-driven route reverses that emphasis. Commercial availability can expose an agency to capabilities it would not have designed itself and can shorten the path to a pilot or service. But the convenience can also conceal a decision. If a chatbot, ranking function, image classifier, or generative component enters a larger product without being named in the original requirement, the agency may not have defined how to test it, what data it may use, or what public impact it could have. The problem is not the vendor’s initiative by itself. It is the absence of a deliberate second look when the proposed capability changes the nature of the service.

The same trade-off appears between custom acquisitions and established contract vehicles. A custom acquisition is suited to a specific mission need that requires substantial agency involvement during performance. It gives the agency a chance to shape the requirements, development process, and terms around that need.