The U.S. public sector ends AI "experiments" and accelerates the adoption of agent-based labor

American public institutions are no longer viewing artificial intelligence (AI) as a “future technology,” but rather as an “immediately deployable tool” that can instantly lighten workloads. Amidst stagnant workforce growth, administrative demands are increasing, coupled with outdated systems and strict regulations, AI is emerging as a core means to boost productivity and achieve modernization.

Google Cloud recently responded to this trend with “Gemini for Government” as a leading initiative. Its core focus is positioning AI not just as an auxiliary function, but as a “platform layer” that enhances overall public sector business productivity and improves aging IT environments.

This shift is confirmed by comments from Google Public Sector CTO Chris Haines on The Cube Research podcast “App Dev Angle.” Haines analyzed that as public agencies integrate AI into actual operations, an “Agentic Workforce” is emerging. He explained that AI is evolving toward becoming “a valuable part of the team” rather than replacing organizations.

Beyond chatbots, evolving into AI that assists civil servants

Over the past two years, corporate AI adoption has often remained at pilot projects, proof of concepts, and chatbot-centric levels. However, in the public sector, this trend is becoming more pragmatic. To handle more work with fewer personnel, gradually integrating AI into existing workflows to improve productivity is being promoted first.

For example, applying AI to administrative document processing, information retrieval, internal automation, and other repetitive, time-consuming tasks to improve daily efficiency for civil servants. This can be seen as an early form of “Agentic Workforce.” It is a structure where humans make the final judgment, with AI acting as a practical collaborator.

Market research also supports this. According to The Cube Research, 46.5% of organizations face pressure to develop applications faster than three years ago. However, workforce expansion cannot keep pace. As a result, AI is viewed as a highly promising substitute to bridge this gap.

The core of AI in public institutions is not performance, but “compliance”

In private enterprises, innovation often precedes regulation, but public institutions are different. Security certifications, data storage locations, personal information protection, and regulatory compliance are not secondary factors but starting points. When adopting AI, the primary concern is not performance but “whether it operates within permissible environments.”

Google explains that to meet these needs, they have adopted a strategy of embedding control mechanisms directly into the cloud platform itself, rather than isolating AI functions in separate closed environments. This way, agencies can satisfy regulatory requirements such as the federal Risk and Authorization Management Program (FedRAMP) and Department of Defense standards, while still utilizing the latest AI models.

This issue is not unique to government. As the European Union’s (EU) “Cyber Resilience Act” and similar regulations increasingly require software and AI to be designed with compliance in mind from the outset, private developers are also facing similar pressures. This indicates that variables influencing AI adoption speed are shifting from technical factors to “internalized compliance.”

Modernization of legacy systems shifting from one-time projects to routine work

The modernization of legacy systems, long considered a stubborn problem in government IT, is also changing with AI’s introduction. Many agencies still operate core systems built decades ago, which in the past were often attempted to be overhauled in large, one-off replacement projects.

Now, approaches such as AI-assisted code cleanup, documentation supplementation, and workflow improvements are gaining attention. Notably, AI has begun to be used to reconstruct long-legacy COBOL-based systems that are difficult to convert.

This aligns with a shift in mindset: viewing modernization as an ongoing process rather than a “project with a finish line.” In the entire application development field, modernization is no longer a goal tied to a specific point in time but a continuous operational principle.

The increasing importance of “choice” not tied to a single AI

Public agencies are increasingly wary of long-term dependence on specific AI vendors or models. Haines emphasizes that the key value for government is “optionality”—the ability to choose. His view is that agencies must be able to select different models based on performance, cost, security, and use case considerations to respond to the rapidly evolving AI market.

Platforms like Google’s Vertex AI aim to meet this need by offering both open-weight models and cutting-edge models simultaneously. This is also significant for developers. In the future, applications are less likely to be tightly bound to a single model and more likely to be designed to interact with multiple models.

Edge deployment expands… AI is no longer just a cloud matter

AI workloads are no longer confined to centralized clouds, which is another important shift. In public settings, due to latency, network disruptions, and operational environment constraints, inference on central servers is often unsuitable. Therefore, the demand for “edge AI”—deploying and adjusting models on-site or in distributed environments—is rapidly growing.

However, this presents new challenges for development and platform operation teams, as they need to jointly manage model lifecycle, governance, and deployment across both central and edge environments. Ultimately, AI is transforming from a simple cloud service into an “architectural layer” that runs throughout the entire application environment.

The trends in the U.S. public sector show a different picture from the traditional view that “governments lag behind private companies in AI adoption.” Due to workforce limitations and the heavy burden of basic service operations, some even assess that public agencies are deploying AI more rapidly in terms of productivity improvement and system modernization. While “Agentic Workforce” remains an evolving model rather than a fully mature concept, AI is clearly moving beyond just adding features to becoming an integral part of labor, platforms, and software architecture itself.

TP AI considerations This article is summarized based on the language model of TokenPost.ai. Some main content of the original text may be omitted, and factual inaccuracies may exist.