Public sector AI will succeed (or fail) based on context 

Public servants today face a double burden: They’re simultaneously charged with running our most important community functions—like disaster preparedness and administering elections—while the technology at their disposal is outdated and ill-fitted to the job. 

The rise of AI has upended how private companies operate, but public servants across agencies lack AI tools designed specifically with government work in mind. 

In a perfect world, public servants could trust mass-market AI. But provisioning critical services requires a high bar. Quickly deploying technology prone to generating inaccuracies is an unacceptable tradeoff for those solving society’s hardest problems. Few of us would be happy to get our mail a day earlier if it meant 10% of our mail never came. The tradeoff is more acute when public servants are working to end homelessness or reinvigorating economic development.  

AI is more accurate and useful for government employees when it’s built atop core data assets like documents and emails and the contextual metadata around those assets—information like who shared documents, when they were shared, and the conversations that surrounded them.  

Public sector AI requires context 

The massive opportunity to empower public servants and improve government operations with functional, reliable AI tools comes not from training larger, smarter models, but ensuring AI has context. 
 

Context is everything because government operations depend upon the local partners, procedures, history, and regulations of a community of practitioners. For AI to work for public servants, it must understand the context well enough to generate accurate information.  

Today’s AI tools fall short because they lack contextual metadata. Without this information, AI is not fit for purpose. Public servants cannot sacrifice accuracy for speed. 

Siloed technology destroys context 

Government work is inherently collaborative. Cybersecurity officials work with state and federal counterparts, and homelessness coordinators work with public health departments. But there is a fundamental mismatch between the collaborative nature of government work and the silos of most technology. 

Today’s AI tools generally serve single organizations, lacking functionality to enable cross-agency collaboration. When FEMA responds to disasters, utilities, hospitals, shelters, and community organizations all play key roles. Public servants coordinate these nongovernment partners, but isolated AI systems can only access information within their own agencies—missing the context that lives across organizations. 

And the work doesn’t happen in siloed agency folders. It happens in email threads, texts, unshared working documents, and view-only, versioned, and immediately outdated shared documents. These disconnected digital workspaces destroy context. But this is a technology problem—what does a context-rich technology look like? 

The government operations tech stack 

Effective government AI must be attentive to the different technology layers that underpin the work of public servants. We can visualize the government operations tech stack in four layers: 

• Layer 1: Systems—The first, foundational, layer comprises the file storage systems: OneDrive, SharePoint, local folders, Outlook, and other repositories. While this is where key information often lives, it is rarely well-organized or accessible to outside partners.  

• Layer 2: Resources—This refers to the resources themselves. Think individual files like memos, spreadsheets, SOPs, and more. While enterprise AI systems can access one organization’s documents, they miss the critical context of how and why these resources were shared, who created them, and what discussions they generated. 

• Layer 3: Coordination— The coordination layer encompasses emails, texts, events, direct messages, and video communications. This is where cross-organization collaboration happens and where ongoing discussions shape decisions. It contains the three sentence email from the 30-year department veteran, who succinctly explained where an internal policy originated, why it was created, and which parts no longer apply. This is institutional knowledge shared in real-time. AI tools without access to the coordination layer are set up for failure.  

• Layer 4: Interface—The interface layer is where public servants make use of the data across layers. And this is where purpose-built AI can make an impact. Government officials should be able to get immediate answers without needing to recall whether information lives in a shared drive, email, video call, or calendar event. And the interface layer doesn’t end with a search — it should enable the next step, whether that’s drafting a policy, connecting with a subject matter expert, or reaching out to partners.  

Atop digital layers are public servants making decisions and taking action. This is where policy meets practice, where coordination becomes execution, and where community needs are met. 

Only context-rich AI can reliably scale public impact 

An AI interface with the full contextual metadata of government operations—the systems, resources, and coordination layers—becomes transformative. An elections official searching for polling center volunteers finds not just the sign-up sheet in their drive, but also the follow-up email from a facilities manager identifying the correct entrance, the text from a sick volunteer needing replacement, and the recent listserv discussion correcting the record about the polling location entrance. AI with this context provides a complete operational picture, not isolated documents that become outdated as soon as they’re created. 

During emergency response, an AI with contextual access can connect FEMA policies with real-time partner communications, community feedback, and operational updates. Instead of just knowing what documents exist, the AI understands who shared critical information, when situations changed, and why certain decisions were made, enabling more effective coordination and faster response times. 

This contextual AI doesn’t just provide information—it provides traceable, auditable insights that public servants can trust and act upon. It connects users not only to the right documents but to the right people and the right conversations, embedded within their specific community and operational context. 

The vision is clear: AI that lives where government work happens, with access to the full collaborative environment across organizations. When deployed with complete contextual metadata, AI can empower public servants to make a bigger impact while maintaining the accuracy and accountability needed. Government operations are fundamentally about coordination and context, and AI must reflect this reality to succeed in the public sector. 

Madeleine Smith is cofounder and CEO of Civic Roundtable.