For decades, build vs. buy wasn't much of a debate for government licensing systems.
Agencies bought a commercial off-the-shelf (COTS) or SaaS platform, adapted their internal processes to fit it, and accepted recurring license fees as the price of avoiding a multi-year custom ground-up software project. That trade-off made sense when a project like that took years and multiple engineering teams to pull off.
AI-assisted development is worth factoring into that calculation now, not because it makes buying the wrong choice, but because it changes what building a purpose-built system costs in the first place. By purpose-built, we mean software designed around one agency's own statutes and workflows, the approach procurement documents often call custom development.
Why Buying Became the Default
Government software built from scratch has a long history of expensive, delayed projects, so COTS and SaaS platforms became the safer default: lower upfront cost, faster time to a working system, and a vendor to call when something breaks. The trade-off was real too. Fitting an agency's processes to a vendor's platform, rather than the other way around, meant workaround friction, recurring per-seat fees that scale with staff and applicant volume, and a roadmap controlled by someone outside the agency.
What's Different Now
Federal agencies spend more than $100 billion a year on IT, and roughly 80% of it goes toward operating and maintaining systems that already exist, according to a July 2025 Government Accountability Office report (GAO-25-107795). That ongoing cost is why the build or buy choice matters so much: whatever an agency picks, it will be paying to run it for years.
AI-assisted development tools can shorten parts of the build timeline that used to make "buy" the obvious default. A widely cited 2024 study of more than 4,800 developers at Microsoft, Accenture, and a Fortune 100 company found that developers using an AI coding assistant completed 26% more tasks than those working without one (MIT, Princeton, and University of Pennsylvania researchers, as reported by IT Revolution, September 2024). That's a meaningful gain, though the research is still catching up. METR, whose 2025 trial found experienced open-source developers took longer with early-2025 AI tools, reported in February 2026 that its follow-up pointed toward a speedup, but said selection effects make the size of that gain hard to pin down. The honest takeaway: AI can speed up pieces of software development. It doesn't remove the need for experienced engineers, good architecture decisions, or disciplined review.
Where a Custom Build Can Make Sense
For licensing systems built around specific statutory requirements (professional licensing boards, permitting, occupational licensure), a few advantages of purpose-built software are becoming accessible to more agencies:
Built to match the statute, not the other way around. Purpose-built logic can mirror specific ordinances and regulatory requirements directly, rather than forcing the agency's process into a generic vendor workflow.
Ownership stays with the agency. The agency or its implementation partner controls the codebase and roadmap, rather than depending on an outside vendor's release schedule and pricing changes.
A different cost curve. Recurring per-seat license fees are replaced by infrastructure hosting and maintenance costs, which can be more predictable as user counts grow, particularly in FedRAMP-authorized cloud environments built for public sector data requirements.
The Trade-offs That Don't Disappear
Faster development doesn't remove the responsibilities that come after launch:
Technical debt still accumulates. Code generated quickly still needs testing, governance, and ongoing maintenance like any other system.
Turnover is still a risk. Thorough, current documentation matters even more when a system is built and maintained by a small team rather than a large vendor's support organization.
Security review isn't optional. AI-generated code can introduce insecure dependencies or logic gaps just as easily as it can accelerate delivery, which makes static and dynamic security testing and human code review non-negotiable steps before anything reaches production.
The Real Question for Agencies
None of this makes buying the wrong answer for every system, and it doesn't make building the right one either. What's changed is the size of the gap between the two options. For licensing systems with heavy statutory complexity or a history of forcing agencies into vendor workarounds, custom development is a realistic option in a way it wasn't a few years ago, provided the agency (or its partner) treats AI-assisted code the same way it would treat any other code: tested, reviewed, documented, and governed.
At The Canton Group, we've spent nearly three decades helping agencies work through exactly this kind of modernization decision, including the procurement rules and oversight requirements that come with it. The tools have changed. The judgment required to use them well hasn't.
