AI Will Write the Government Software. Who Writes the Spec?
Civic Tech can be serious work, but a touch of humor keeps it human.
//
Every government IT contract begins with a scope of work.
And ends with a change order.
//
A government technology director described her situation recently. She had two options in front of her: a $4 million custom software contract, familiar, defensible, understood by her procurement office, and a newer approach where AI generates the code from a plain-language specification her team writes and owns. The second option was faster and cheaper. But her procurement office had no framework for it. Her legal team had never reviewed a contract structured that way. And her program staff, the people who actually know what the system needs to do, had never written a specification document in their lives.
As she talked, I noticed she kept framing this as a build-versus-buy question. Which option was safer. Which vendor had better references. Which fit the existing procurement vehicle.
But the more I listened, the more I suspected she was solving the wrong problem. If AI can generate functional code in minutes from a well-written specification, what’s the asset that actually matters? Not the software. The specification. And that one shift changes the entire frame of the decision.
If you zoom out, the pressure AI is putting on government technology falls into three levels: how individual agencies think about what they own, how procurement frameworks need to adapt, and what it means for the civic tech field’s twenty-year mission of sharing code. Here are three ways that shift is already playing out.
1. Own the Spec, Own the System
The first transformation happens at the asset level. For decades, the thing government agencies paid for, and often struggled to own, was software: code written by vendors, licensed on terms the agency didn’t fully control, dependent on vendor knowledge to modify or extend. AI changes the economics of that relationship.
As AI coding tools mature, software becomes increasingly regenerable. The Civic Innovations blog captured this in January 2026 with the concept of “disposable software,” the idea that when an AI coding agent can regenerate a functional component from a prompt, the incentive to maintain and refactor code over years diminishes. What holds its value is the durable layer: the specification, the policy logic, the documented edge cases, the plain-language description of what the system must do and for whom.
Agencies that own that layer own their systems, regardless of which tool generates the implementation. Agencies that outsource the specification are still dependent, just on different vendors than before.
The outcome is a fundamental reorientation of where government technology investment should go: from buying software to building the organizational knowledge that makes good software possible to specify in the first place.
2. The Procurement Framework Hasn’t Caught Up
The second transformation needs to happen at the institutional level, and right now it is moving slowly.
Government procurement is built around software as a product: scope it, buy it, maintain it. It has no established framework for software as a regenerable output of a specification the agency writes and owns. Contracting officers don’t know how to evaluate a spec-first engagement. Legal teams don’t know what to do with a contract whose core deliverable is a living document rather than a deployed application.
The Legacy IT Reduction Act of 2026, bipartisan legislation just introduced in Congress, would require agencies to inventory every legacy system and develop five-year modernization plans. That’s a useful step. But the bill doesn’t address this shift at all. It is solving the 2015 version of the problem: how to plan migrations, not how to own what comes after them.
The agencies getting ahead of this are changing the question they bring to procurement. Instead of “which vendor do we buy this from,” they are asking “what does this system need to know, and how do we document that knowledge so we own it regardless of who builds the implementation.” That reframe removes the leverage that has made vendor lock-in so durable for so long.
3. What the Civic Tech Field Actually Needs to Share
The third transformation reaches the field itself. The “build once, share widely” vision has driven open-source government technology for two decades. The assumption was that the bottleneck was code, and sharing it would lower the cost of building better public services.
If AI can regenerate implementations cheaply from a good specification, that assumption needs revisiting. The bottleneck was never the code. It was always the knowledge embedded in the specification: the policy logic, the user research, the failure modes, the edge cases that took years of operational experience to discover.
The organizations building that knowledge commons, documenting what worked, what failed, and why the system does the thing it does, are the ones that will actually move the field. Sharing repositories without sharing the knowledge behind them is solving a problem that AI is already solving on its own.
The Spec Is the New Source Code
The most important question in government technology right now is not which AI tools agencies are deploying. It is whether agencies are building the organizational capacity to tell those tools what to build.
When programs own their specifications, vendors compete on execution rather than institutional knowledge. When procurement frameworks catch up, agencies can move faster without surrendering control. And when the civic tech field starts sharing specifications and policy logic alongside code, the compounding effect reaches everyone.
The question isn’t whether AI will write government software. It will. The question is who holds the pen on what gets written, and whether that person works for the agency or the contractor.
The Civic Tech Daily Team
Questions or thoughts? Comment below, email us at hello@civictechdaily.com, or visit www.civictechdaily.com. Follow us on Substack and X @CivicTechDaily.
Civic tech is in our daily lives. Thank you for following Civic Tech Daily, where we bring you closer to the policies, tools, and people shaping the future of civic technology.