U.S. Immigration and Customs Enforcement (ICE) is exploring an agentic software factory that would use AI agents to handle significant portions of software development, according to a September 30 FedScoop report. 

The agency published a request for information (RFI) on Sam.gov last Friday seeking industry input on expanding its existing Secure Technology Environment for Large Language Agents (STELLA) platform.

ICE envisions the software factory using AI agents for planning, implementation, testing and review across the software development lifecycle. The proposed system would include an enterprise control plane for governance, architecture, infrastructure, shared services and cost reporting, along with an orchestration layer for managing agent tool access, human escalation and stop conditions.

The RFI reflects ICE’s plans for future capabilities rather than an existing implementation. ICE said its intent is to expand STELLA and existing development capabilities while keeping generation, assurance and authorization separate and preventing autonomous components from expanding their own permissions or approving their own production releases.

ICE Wants Agents Under Centralized Controls

STELLA provides a secure environment for developing, testing and deploying AI use cases. ICE previously said the platform includes authentication and authorization, security and code scanning, AI guardrails, logging and analysis, and access to large language models.

ICE said at least a dozen production-ready AI applications are already live on STELLA. Under the proposed software factory, a vendor would help develop and deploy additional AI applications and integrate frontier and specialized AI models, according to the report.

The agency also expects the prospective vendor to provide workforce training on the software factory and its safe use. ICE said the vendor could assume continuous monitoring responsibilities for its enterprise data and analytics ecosystem while enabling the software factory, the FedScoop report states.

Agentic Development Brings Data Governance Into the Architecture

For chief data officers (CDOs), the proposal highlights how agentic AI can make data governance part of the architecture supporting software development. Agents that can use tools, interact with multiple systems and perform development tasks require clear rules for what data and resources they can access.

The proposed orchestration layer would give ICE a way to define tool access, human escalation and stop conditions. The enterprise control plane would provide another layer for governance and cost reporting. Together, those components show how organizations can build controls around agents before giving them broader operational responsibilities.

ICE’s planned continuous monitoring of its enterprise data and analytics ecosystem also makes data visibility a central part of the effort. For CDOs, that means tracking which agents interact with organizational data, what permissions they receive and how those permissions are governed as AI use expands.

The approach also reinforces the importance of separating AI development from authorization and assurance. For organizations deploying agentic systems, AI governance can provide a framework for defining access, accountability and oversight as agents move into production.