AI Will Write the Government Software. Who Writes the Spec?

Key Takeaways

• Specifications become the primary asset, not the generated code
• AI enables rapid code regeneration, reducing long‑term maintenance costs
• Procurement rules must evolve to contract for living specification documents
• Civic tech should share knowledge commons alongside open‑source code
• Agency staff need training to author clear, executable specifications

Pulse Analysis

The rise of generative AI coding tools is turning the traditional software procurement model on its head. For decades, agencies have paid for code that they could barely modify, relying on vendors for updates and bug fixes. Today, a well‑crafted specification can be fed into an AI engine that produces functional code in minutes, making the specification the true source of value. This "disposable software" paradigm means that the durable knowledge—policy logic, edge‑case handling, and user requirements—must be captured in a living document that agencies can maintain independently of any vendor.

At the institutional level, procurement offices are scrambling to keep pace. Current contracts are built around buying a finished product, with clear scopes and change orders. AI‑first engagements, however, deliver a mutable output that evolves as the specification does, leaving contracting officers without a framework for evaluation, risk assessment, or performance metrics. Recent legislative efforts like the Legacy IT Reduction Act of 2026 focus on inventorying legacy systems but ignore this emerging shift. Agencies that proactively rewrite their RFPs to ask "what knowledge must the system embody" rather than "which vendor will build it" are gaining strategic leverage and avoiding future lock‑in.

For the broader civic‑tech ecosystem, the implication is profound. The long‑standing mantra of "build once, share widely" assumed code was the bottleneck. AI removes that barrier, exposing the real scarcity: high‑quality specifications. Communities that curate and openly share specification repositories—complete with policy rationales, user research, and failure analyses—will accelerate innovation far more than code‑only libraries. Building internal capacity to author and maintain these specifications becomes a competitive advantage, ensuring agencies retain control while benefitting from AI’s speed and cost efficiencies.