In November 2018, people visiting the Secretary of State office in Troy waited an average of 112 minutes before receiving assistance. By 2021, Michigan reported that visitors there were completing their entire office visits in about 20 minutes. That's the kind of improvement people remember.
The state redesigned a familiar service around the people using it.
AI is now changing what capable government technology teams can build. In Alberta, officials faced a vendor bid of approximately $54 million to replace an aging system. A team of public servants used AI development tools to build two replacement systems, projecting about $2.6 million in total costs. Those are the province's estimates, not audited savings, but they demonstrate why government needs the expertise to decide what to build and what to buy.
The first two pieces in this series argued for building new industries around Michigan's strengths and giving startups better access to customers. State government has an opportunity to advance both ambitions while improving its own services.
And if that work saves money, Michiganders should see where the benefit goes.
Make Michigan government a builder and a better customer
Michigan already has an AI Core Team and a technology department. I'd build on that foundation with experienced AI engineers and product leaders working alongside agency staff, able to develop focused applications, maintain them and challenge vendor proposals. Specialist firms will still be needed, but the state should have enough expertise to decide what to build, what to buy and what a fair price looks like in the AI era.
Pennsylvania's Department of Human Services tackled a familiar problem: people submitting unreadable documents for public benefits, forcing caseworkers to chase paperwork instead of helping residents. Working with the state's digital service team and outside technology partners, it introduced AI to check documents as they were submitted. In a two-month pilot covering approximately 12,000 documents, the state reported an 80% reduction in illegible submissions and more than 700 hours of staff time saved. Caseworkers retained responsibility for eligibility decisions. That's the kind of improvement Michigan should pursue: using AI to eliminate unnecessary work while giving public employees more time to serve people.
Michigan has already debated how to build this capability. In May, HB 5504 and HB 5505 passed the House and remain pending in the Senate. HB 5504 would create a digital service office within DTMB, operating independently of its director and able to block or redirect certain technology projects. HB 5505 would establish a legislative oversight office with authority to intervene in agency projects. A DTMB representative opposed the bills in a House committee.
The bills reflect a real concern about government technology delivery, but competing authorities could make responsibility harder to pin down when projects go wrong. I'd strengthen engineering and product capabilities within the executive branch and give legislators clear oversight of spending and results. That would combine stronger delivery with a single accountable chain of command, rather than another layer of approval.
Getting that structure right matters because Michigan government should do more than become a better overseer of outsourcing. A capable internal technology team could take on more work itself and become a much smarter buyer when outside expertise is needed. Rather than defaulting to large contracts with national firms, the state could commission more focused work that qualified Michigan companies, including startups, have a genuine opportunity to win.
A Michigan company that solves a difficult problem for one agency could turn that experience into a product sold to other governments and businesses. That's the connection to the first two pieces in this series: use the institutions we already have to help build new industries here. And as we bring more work inside government, we should measure how much expertise and employment we're creating in Michigan, recognizing that national vendors also employ Michiganders. The opportunity is to deliver better public services while building a technology industry that can compete well beyond our state.
Make sure employees benefit from the change
Michigan is already thinking seriously about what AI will mean for workers. The state's 2026 progress report on its AI Workforce Plan found that roughly one in ten Michigan workers holds a job highly exposed to AI, particularly in administrative and office roles. The state has also expanded access to AI and digital skills training through all 16 Michigan Works! agencies.
That's a meaningful start. I'd like Michigan's own government transformation to put those principles into practice.
I've led large transformations at public companies where employees had to learn new systems while keeping the old ones running. It puts an immense strain on people, especially when employees are uncertain about what happens to their jobs. An ambitious transformation can be well-guided from a technology perspective, but can ultimately fail because we underestimate the change management required to have the new approach take hold.
We should make a clear commitment before major AI deployments begin. Employees should help redesign the work, with training during paid working hours and opportunities to move into new responsibilities. During an agreed transition period, no employee should lose their job solely because an AI pilot projects fewer hours of work. We should use retirements, vacancies and redeployment to manage longer term changes, while bringing more technical expertise and work into Michigan government.
Pennsylvania established a labor-management group to give state employees a voice in how AI is introduced into government operations. Michigan should build on that model, giving workers a real role in redesigning their work and a fair opportunity to benefit from the changes.
The opportunity is to build a government where employees have better tools, stronger skills and more rewarding work. We should be able to improve services substantially without asking public employees to bear the cost of that progress.
Never let automation become a barrier to care
That same commitment has to extend to the people government serves.
Michigan has experienced the consequences of getting automated decisions wrong. Its MiDAS unemployment system falsely accused thousands of residents of fraud, leading to wrongful seizures of wages and tax refunds and, ultimately, a $20 million settlement. Automation can magnify errors as easily as it eliminates paperwork.
Michigan is again using AI in benefits administration. State officials report less manual work and say staff retain eligibility decisions. But the ACLU of Michigan has questioned a tool that reviews SNAP payments, asking about accuracy and protections against wrongful denials. A human in the process is not enough if that person cannot question the system or explain its decision.
AI should make applications easier to process and legitimate claims faster to approve. But decisions that could deny someone health care, income or essential benefits need meaningful human review, a clear explanation and an accessible appeal.
Rep. Carrie Rheingans and colleagues have introduced HB 4536 and HB 4537 addressing AI-driven insurance and medical assistance denials. The concern is important, though the bills could better distinguish helpful AI assistance from automated decisions. Colorado enacted legislation this year permitting AI-assisted reviews while requiring qualified human oversight of medical-necessity denials. Michigan should pursue that balance: use AI to improve access without making it another barrier to care.
Give Michiganders a share of the savings
If AI lets Michigan deliver better services while spending less on outside technology contracts, the public should be able to see where those savings go.
I'd call that a Michigan AI Dividend.
The idea isn't to promise everyone a check. It is to create a transparent account of the financial gains from better technology delivery and return those gains to the people of Michigan.
That starts with measuring the right things. Saving an employee two hours a day is a valuable productivity improvement, but it doesn't necessarily mean the state has saved money. We'd need to compare the full costs of an old system and its replacement, including technology, employees, training, security and ongoing maintenance. Only independently verified net savings should count. And savings supposedly achieved by denying people benefits should never qualify.
Washington law allows agencies to reinvest certain qualifying savings in one-time improvements such as employee training and technology. Michigan would need its own legislative framework, transparent accounting and protections for funds dedicated to essential services.
I'd give agencies and employees a share of verified net savings to reinvest in better tools, training and services. Only if savings prove substantial and recurring should we consider a broader return through priorities such as health care, roads or tax relief. The first dividend is better service; any financial dividend must be earned and visible.
The bigger opportunity is to make the benefits visible. Imagine Michigan becoming known for government services that work exceptionally well, public employees who are equipped to do their jobs, and a growing technology industry that can sell what it builds here across the country.
That's how the three pieces come together. We can build new industries around Michigan's existing strengths, give promising companies customers willing to invest in their success, and use government itself to demonstrate what responsible AI transformation can accomplish.
I'd like Michiganders to see that progress in the services they rely on and in the careers available to them. And when their government saves money, they should be able to see exactly what they received in return.
