Government software development is subject to a complex web of regulations that mandate strict compliance with security, accessibility, interoperability, and documentation standards. These regulations increase the cost of development and slow innovation, but they serve as important mechanisms for ensuring higher-quality, more secure, and more maintainable software systems. LLMs can accelerate software development by rapidly generating code, documentation, and test cases, significantly reducing the time needed for routine or boilerplate tasks. Their flexibility allows developers to explore multiple design alternatives or adapt solutions to evolving requirements with minimal additional effort. Using LLMs to write software is challenging for government organizations because LLMs are based on probabilistic outputs and have not yet proven to generate software that meets the standards for reliability, maintainability, security, evolvability, and auditability. Governments have some advanced software design patterns that are required to meet specialized requirements including, but not limited to: Very large-scale data, large number of users, real-time systems (with performance guarantees), operation on air-gapped networks, safety standard compliance, classified data handling and communication, constrained hardware for computation and networking, and optimization across distributed sites. These contexts highlight important boundaries for LLM-assisted development, where traditional human expertise, specialized tooling, and domain-specific methodologies remain essential components of the software engineering process.