Government Modernization Is Not a Software Problem

---

Issue Areas

---

A Reuters investigation published in August 2025 reported that two nearly complete Pentagon HR software projects—despite 12 years of work and well over $800 million in combined cost— were still at risk of being scrapped and redone.¹ This is yet further evidence that government modernization is not primarily a software (or hardware) problem. It is an implementation problem.

The recent reaction to state dysfunction is understandable: bring in more engineers, buy better tools, automate more workflows, and force agencies off legacy systems. Government does need more technical talent. But the federal record shows that talent and tools are not enough. In its 2025 high-risk update, GAO said agencies still face three core challenges: weak oversight of IT portfolios, weak acquisition and development practices, and weak federal IT capacity and capabilities.² Until those are fixed, agencies will continue to produce systems that arrive late, run over budget, slip schedules, and contribute too little to mission outcomes.

This should not be surprising. The broader productivity literature has been making the same point for decades. Brynjolfsson and Hitt argued that the value of IT depends heavily on complementary organizational investments, and that many benefits are hard to measure if one looks only at the technology itself.³ Bresnahan, Brynjolfsson, and Hitt likewise found that the gains from IT are greater when it is paired with organizational change rather than dropped into an unchanged workplace.⁴ Technology matters, but institutions have to change around it.

The public-sector evidence points in the same direction. Electronic procurement in India and Indonesia improved quality, but did not reduce prices.⁵ Electronic tax filing in Tajikistan reduced firms’ compliance costs by 40 percent and changed the pattern of tax and bribe payments in ways consistent with lower opportunities for collusion or extortion.⁶ Biometric smartcards in India made welfare payments faster, more predictable, and less corrupt without reducing access.⁷ The lesson: technology can improve state capacity when it is embedded in an administrative system that is redesigned to use it well.

That is why “modernization” should be treated as a strategic implementation problem, not a procurement event. GAO’s 2025 testimony on agency reform makes the point directly: effective reform requires a combination of people, processes, technologies, and other critical success factors. It also stresses that reforms should be anchored in clear outcome-oriented goals and performance measures, developed with employees and stakeholders, and justified with data and evidence.⁸ In other words, buying software or hardware is the easy part (and procurement has its own challenges). Reorganizing work around it is the hard part.

Once you see modernization this way, many government failures become easier to explain. Leaders can order inventories, pilots, dashboards, and status reports. They can demand AI adoption. But those are easier to observe than whether an agency has actually redesigned a workflow from end to end. And the federal government is already far beyond the question of whether adoption is happening. Federal agencies are required to inventory public AI use cases, and the official 2024 consolidated inventory reported 41 agency submissions and 2,133 AI use cases. That is evidence of activity. It is not, by itself, evidence of transformation.

An agency can accumulate AI use cases while leaving the underlying production process mostly untouched. It can also make work more cumbersome by layering new tools onto old review chains, fragmented vendor systems, and compliance rituals that nobody has redesigned. OECD work on public-sector innovation warns that high-accountability cultures and fear of failure often distort innovation itself: projects may be killed early, allowed to drift into “too big to fail” status, or evaluated in ways that reward activity rather than outcomes. Excessive control, OECD notes, can itself stifle innovation.⁹ That is exactly the environment in which symbolic modernization tends to beat substantive modernization.

This is the real modernization dilemma. If visible mistakes are punished more than invisible stagnation, then agencies should be expected to prefer safer, more legible forms of compliance over deeper process redesign. They will launch pilots rather than rewrite workflows. They will count tools rather than measure cycle times. They will add systems rather than remove steps. That response is not necessarily irrational. It is what incentive-compatible behavior looks like inside a high-oversight bureaucracy. The practical question, then, is not whether government will use AI. It already is. The question is whether agencies are being pushed to use it for actual process improvement.

That is where the research and policy agenda should move. We should spend less time asking whether agencies have “adopted AI” and more time asking whether an end-to-end workflow became simpler, faster, cheaper, or more reliable. Which audit structures deter box-checking without deterring experimentation? Which performance measures track outcomes rather than procurement activity? When should agencies be allowed to keep or redeploy savings generated by modernization instead of being punished for producing them? And what kinds of phased rollout, leadership protection, and stakeholder buy-in make deeper reform politically survivable? GAO’s own framework points in this direction: start with goals and outcomes, involve frontline users, use evidence, and measure reforms against mission results rather than rhetoric.

Government needs engineers. It also needs managers, budget officials, auditors, procurement staff, and political overseers operating under rules that reward simplification and outcomes rather than theater. Without that, modernization will remain a cycle of demos, dashboards, and expensive disappointments. With it, better tools can change how the state works.