---
title: "Government AI Agents: Build Once, Share Across Agencies"
slug: "government-ai-agents-build-once-share-across-agencies"
author: "Miguel Amigot"
date: "2026-11-19 12:00:00"
category: "Premium"
sector: "government"
topics: "government AI agents, OMB M-25-21, AI code reuse federal agencies, federal AI use case inventory, cross-agency AI sharing, MCP government, AI agent reuse, government AI procurement, source code ownership"
summary: "OMB's M-25-21 tells federal agencies to share and reuse AI code and models, yet agencies reported more than 3,600 AI use cases in 2025 and most of those with development details involve contractors. How an agency builds an AI agent once and lets other agencies copy it or call it, and the contract term that decides whether it can."
banner: ""
thumbnail: ""
linkedin: |
  Many agencies face similar AI problems: answering the public, triaging cases, summarizing documents. Each one can end up building its own agent.

  The policy already points the other way. OMB's M-25-21, issued April 3, 2025, tells agencies to proactively share their custom-developed AI code, including models and model weights, whether they built it or bought it.

  One of its four exceptions is the one that matters: an agency can be "prevented from doing so by a contractual obligation."

  That matters because Brookings found about 63% of 2025 federal AI use cases with development details were built by contractors, alone or with in-house staff.

  So reuse is decided at signing. Two ways to share an agent once you can:
  → Copy it into another agency's own organization, where it can be adapted
  → Call it over MCP, so the agency that built it keeps running it

  With ibl.ai you own all the code and the data, and agents copy across organizations and answer over an MCP endpoint. Build once, share across agencies.

  #iblai #AgenticAI #EnterpriseAI #GovTech #AIAgents #MCP
faqs:
  - question: "Does federal policy require agencies to share AI code and models?"
    answer: "Yes, with exceptions. OMB Memorandum M-25-21, issued April 3, 2025, requires agencies to proactively share across the federal government their custom-developed AI code, including models and model weights, for AI applications in active use, whether developed in-house or procured. Sharing is excused where law restricts it, where it creates an identifiable risk, or where a contract prevents it."
  - question: "What is the difference between copying an AI agent and calling it?"
    answer: "Copying puts an instance of the agent in the receiving agency's own organization, where its team can adapt and run it. Calling leaves the agent where it was built and lets another program use it over a standard interface such as MCP, so the owning agency keeps maintaining one agent."
  - question: "How does a contract stop an agency from sharing an AI agent?"
    answer: "M-25-21 lets an agency skip sharing when it is prevented from doing so by a contractual obligation. If the vendor keeps the code, model or agent configuration, the agency may have nothing it is allowed to hand to another agency, so the reuse question is settled at contract signing."
---

## The Short Answer

**Government agencies avoid rebuilding the same AI agent when one agency builds it once and the others either copy it into their own organization or call it over a standard interface, which is the reuse OMB's M-25-21 already asks for. On ibl.ai an agent copies across organizations and answers over an MCP endpoint, and you own all the code and the data, so a vendor contract is not what stands in the way.**

Federal agencies reported more than 3,600 AI use cases in 2025, and the policy telling them to reuse rather than rebuild is already on the books. What can decide whether reuse happens is a clause in the contract that built the agent.

This post covers what the policy says, what stands in the way of reuse, and the two practical ways to share an AI agent between agencies.

## What does OMB M-25-21 say about sharing AI code and models between agencies?

[OMB Memorandum M-25-21](https://www.whitehouse.gov/wp-content/uploads/2025/02/M-25-21-Accelerating-Federal-Use-of-AI-through-Innovation-Governance-and-Public-Trust.pdf), issued April 3, 2025, tells agencies to protect taxpayer dollars from duplicative spending and to reuse data, models, code and assessments of AI performance.

Its section on reuse goes further. Agencies must proactively share across the federal government their custom-developed AI code, including models and model weights, for AI applications in active use, whether the agency developed it or procured it.

The memo lists four exceptions: sharing is restricted by law, it creates an identifiable risk to security, privacy or the public, it creates a risk to agency operations, or the agency "is prevented from doing so by a contractual obligation."

It also asks a Chief AI Officer Council to promote shared templates, formats and technical resources. The direction is clear: build once, then reuse.

## Why is AI agent reuse across government agencies hard to coordinate?

Government AI use is growing fast, which makes coordination harder. OMB's [2024 consolidated inventory](https://github.com/ombegov/2024-Federal-AI-Use-Case-Inventory) listed 2,133 AI use cases from 41 agency submissions as of January 23, 2025.

A year later the count was far higher. [Brookings' review of the 2025 inventories](https://www.brookings.edu/articles/assessing-the-state-of-ai-adoption-across-the-federal-government/) found 41 agencies documenting more than 3,600 use cases, 69% above 2024 and five times 2023.

Generative AI grew fastest. GAO counted generative AI use cases at 11 agencies rising from 32 in 2023 to 282 in 2024, roughly nine-fold, in [GAO-25-107653](https://files.gao.gov/reports/GAO-25-107653/index.html).

The inventories show overlap even inside one agency. Brookings notes that some agencies listed the same tools multiple times for different offices, sometimes with different risk classifications.

GAO also records a structural limit: VA officials noted that existing privacy policy can prohibit information sharing with other agencies. Sharing an agent has to work without sharing the data behind it.

## Who decides whether a government AI agent can be reused by another agency?

Often the contract does. Brookings found that around 63% of 2025 use cases with development details were developed by contractors, exclusively or with in-house resources.

By the time use cases reach full deployment, Brookings found, contracting is involved in 72% of them.

Put that next to M-25-21's exception for a contractual obligation. If the vendor keeps the source code, the model weights or the agent's configuration, the agency may have nothing it is permitted to hand to another agency.

That is why reuse is decided at signing, not after deployment. A contract that gives the agency the code and the data turns M-25-21's sharing requirement into something it can actually do.

The clauses that make ownership real are covered in [Government AI Procurement Is Moving From Vendor Selection to Architecture Ownership](/blog/government-ai-procurement-architecture-ownership). This post is about what an agency does with an agent once it owns one.

## How can one government agency share an AI agent with another agency?

There are two practical patterns, and they suit different situations. The table compares them.

<table style="width:100%; border-collapse:collapse; margin:1.5rem 0; font-size:0.95rem;">
  <thead>
    <tr style="background:#f5f5f0; border-bottom:2px solid #2175C5;">
      <th style="text-align:left; padding:0.75rem; color:#5f6368;">Question</th>
      <th style="text-align:left; padding:0.75rem; color:#5f6368;">Copy the agent</th>
      <th style="text-align:left; padding:0.75rem; color:#5f6368;">Call the agent over MCP</th>
    </tr>
  </thead>
  <tbody>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Where does it run?</strong></td>
      <td style="padding:0.75rem;">In the receiving agency's own organization</td>
      <td style="padding:0.75rem;">Where the building agency runs it</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Who maintains it?</strong></td>
      <td style="padding:0.75rem;">Each agency maintains its own copy</td>
      <td style="padding:0.75rem;">One agency maintains one agent</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Can it be adapted?</strong></td>
      <td style="padding:0.75rem;">Yes, to local rules, data and wording</td>
      <td style="padding:0.75rem;">Only by the owning agency</td>
    </tr>
    <tr style="border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>Best fit</strong></td>
      <td style="padding:0.75rem;">Same problem, different data or policy</td>
      <td style="padding:0.75rem;">Same problem, same answer for everyone</td>
    </tr>
    <tr style="background:#f0f9ff; border-bottom:1px solid #e5e7eb;">
      <td style="padding:0.75rem;"><strong>On ibl.ai</strong></td>
      <td style="padding:0.75rem;">Cross-organization agent copy from the auth app</td>
      <td style="padding:0.75rem;">The agent-chat MCP endpoint runs a full agent turn</td>
    </tr>
  </tbody>
</table>

Copying fits when two agencies face the same problem under different rules, such as two benefits programs with different eligibility criteria. The second agency starts from a working agent instead of a blank page.

Calling fits when the answer should be the same everywhere, such as one authoritative agent on a shared regulation. Every agency gets the same answer, and a correction is made once.

## How does ibl.ai let agencies copy and call each other's AI agents?

ibl.ai shipped both patterns in the week of November 6 to 13, 2026, as recorded in the [platform update for November 13, 2026](/updates/platform-update-2026-11-13).

**Copy.** An agent can be copied into another organization from the auth app, and the organization list is searchable. For a government deployment, each agency or program can be its own organization on one platform.

**Call.** The agent-chat MCP tool is reachable at `/mcp/agent-chat/<agent_id>/`, so it works as a custom connector in Claude Desktop or claude.ai with only an `Authorization` header.

Each MCP call runs a full agent turn, using the agent's system prompt, tools, trained documents and memory, and passes moderation, spend caps and credits. An optional session id lets a client continue the conversation.

Because the platform is model-agnostic, a shared agent is not tied to one model vendor. The agency that copies an agent can run it on a different LLM than the agency that built it.

## What should an agency check before it shares an AI agent with another agency?

Start with the four M-25-21 exceptions, because they are the real limits. Confirm that no law, security or privacy risk, operational risk or contract term prevents the sharing.

Then separate the agent from its data. The VA privacy concern GAO recorded, that existing privacy policy can prohibit sharing with other agencies, is a reminder that an agent's instructions and tools can travel while the records behind it stay home.

Finally, choose copy or call deliberately. Thousands of reported use cases leave plenty of room for both: copy where rules differ, call where one answer should hold everywhere.

ibl.ai is family-owned and operated from New York, NY, a U.S.-headquartered, domestically-owned long-term partner for agencies that want to own their agents. More on deployments for public-sector buyers is on the [government solutions page](/solutions/government).

## Why does owning the AI stack matter?

**ibl.ai is the agentic AI platform where you own all the code and the data. You self-host the entire stack inside your own perimeter, run it model-agnostic across any LLM and switch anytime, and pay by usage with no per-seat pricing — so you can deploy anywhere: your cloud, on-premise, GovCloud, or fully air-gapped.**

- **You own all the code and the data.** Full source code under a perpetual license, running on your infrastructure. Not API access to someone else's platform — the stack itself is yours.
- **Model-agnostic.** Run any LLM — Claude, GPT, Gemini, Llama, Command, or your own fine-tune — and switch providers without rewriting the platform.
- **No per-seat pricing.** Usage-based billing against a budget cap you set. Cost tracks what your organization actually uses, not how many people you employ.
- **Deploy anywhere.** Your cloud, your VPC, on-premise, GovCloud, or a fully air-gapped network with no outbound connectivity.

1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University.

ibl.ai is family-owned and operated from New York, NY — a U.S.-headquartered, domestically-owned long-term partner, not a vendor that sells licenses and moves on.

[Explore sovereigngovernment.ai](https://sovereigngovernment.ai/)
