It’s not every day that a government meeting involves someone demonstrating the AI tool that literally just created the agenda for that very meeting. But here we are in 2025, and the meta moments keep coming.
This week’s Digital Innovation Task Force meeting covered everything from statewide collaboration efforts to the eternal question of whether AI will eat our data (spoiler: it’s not the AI we need to worry about – it’s the quality of our data feeding it). We also welcomed Kelly Adkins from the Health Department, who’s diving headfirst into the delightful world of digital systems assessment. Godspeed, Kelly.
Building Bridges: The Statewide AI User Group Takes Shape
Our little survey for Michigan counties interested in AI has grown to 16 respondents – all county folks so far, which means we need to get this out to more organizations. The responses reveal what you’d expect: everyone wants better communications and website management, followed closely by automation and policy development.
Here’s the interesting part: most respondents say they’re “not using AI at all” officially, even though we know about half of county employees are already using AI tools during their work week. It’s the classic gap between “we don’t have an AI program” and “wait, what do you mean I can’t use ChatGPT for this report?”
We’re planning a January virtual kickoff event for everyone who’s interested – a roundtable to figure out what this statewide collaboration should actually look like. If you’re in a Michigan county and haven’t taken the survey yet, visit our Survey’s and Results page.
Cross-County Collaboration: Two Counties Are Better Than One
The proposed meeting model with St. Joseph County is moving forward, and their task force is enthusiastic about it. Here’s how it’ll work: one Van Buren County meeting per month stays in-person for our internal county business. The second monthly meeting goes virtual and combines with St. Joseph County, focusing on broader topics that apply to multiple counties.
St. Joseph’s group includes judges, elected officials, and even council members occasionally, so these joint sessions should generate some excellent cross-pollination of ideas. The virtual format also opens doors for other counties and organizations that have been asking to participate. We’re even discussing creating an FAQ document for new participants – you know, so we don’t spend half the meeting explaining what DITF stands for.
Our next meeting might be the test run for this virtual format, possibly with St. Joseph County joining. We’ll see how it goes. Virtual meetings: because democracy shouldn’t require you to scrape ice off your windshield at 8:30 AM.
When AI Meets FOIA: Let’s Let Someone Else Figure This Out First
FOIA (Freedom of Information Act) compliance and AI-generated content is complicated. Retention policies, what counts as a record, how to handle AI conversations – it’s a legal maze. Fortunately, St. Joseph County has assembled a specialized subgroup with FOIA experts to work through these questions, so we’re going to follow their lead rather than reinvent this particular wheel.
Meanwhile, Drake solved a more immediate problem: people kept choosing the wrong FOIA request form on the county website. Sheriff’s Department? Admin? Who knows! The solution: a simple routing system that asks a couple of questions and guides users to the correct form automatically. No more inefficient back-and-forth trying to redirect requests to the right department.
The system is expandable too – it’ll work for routing different types of patrol service requests and potentially other form-based workflows. Sometimes the best AI application is just making sure people end up in the right place to begin with.
Courts and AI: Progress Through Education
Emma and Liz shared updates from the National Center for State Courts AI webinar series, which continues to be a treasure trove of practical guidance. One major takeaway: instead of expecting everyone to become prompt engineering experts, organizations should build prompt libraries – pre-written, effective prompts for common tasks. Give people the tools, not just the training.
The NCSC also maintains role-based learning resources with job aids specifically designed for attorneys, judges, administrators, court reporters, clerks, and translators. Why build training materials from scratch when you can start with templates from people who’ve already done the work?
Their AI sandbox is worth checking out too – a practice environment where you can experiment with AI tools without worrying about data being used for training. The newest feature is a data extraction tool, which connects nicely to this meeting’s recurring theme: it’s not the AI that’s the problem, it’s getting clean, structured data for the AI to work with.
One policy question raised by the webinar that we hadn’t considered: should employees use their work email or personal email when accessing AI tools? This seemingly small distinction has significant implications for data ownership, liability, and risk. Consider it added to the policy review list.
There’s still resistance to AI adoption in court environments – people are scared, and that’s understandable. But as Emma and Liz emphasized, the webinars make clear that courts can absolutely work with AI while maintaining confidentiality and CJIS compliance. Judge Metzger’s upcoming probate chatbot demonstration might help turn some of those concerns into curiosity.
Drake’s Demo Corner: Skills – The Automation You Didn’t Know You Needed
And now for the meta moment we’ve all been waiting for: Drake demonstrated the skill he built to automate DITF meeting workflows. Yes, the same skill that generated today’s agenda. We’re through the looking glass, people.
Skills are structured instruction sets that extend Claude’s capabilities for specific workflows. Unlike simple prompt libraries, skills encode entire processes: file structures, tool usage patterns, multi-step workflows, and even formatting requirements. Drake’s DITF skill handles agenda generation, summary creation from transcripts, blog post writing (you’re reading the output right now), and featured image generation.
The brilliance is in the automation: Drake can now say “generate the summary for yesterday’s meeting,” and Claude automatically finds the right folder, reads both the agenda and transcript, matches the structure exactly, and produces a comprehensive internal summary. Then it transforms that summary into a public-facing blog post with personality and accessibility for readers who don’t need to know every internal detail.
Skills can even handle real-world complexity. Drake discovered today that the skill’s hardcoded file paths didn’t work on his work laptop, so the solution is to make the skill ask which computer he’s using and maintain configured paths for each environment. Skills evolve.
The architectural insight here is powerful: skills turn “here’s how I want you to help me with X” into persistent knowledge that triggers automatically when relevant. No more explaining the same workflow in every conversation. No more hoping the AI remembers your preferences. Just consistent, repeatable automation for specialized tasks.
And yes, skills can be exported to other AI systems like ChatGPT, though you’d need to manually include them in each conversation rather than having them automatically available. The possibility of using ChatGPT’s memory feature to store skills was raised – an interesting workaround worth exploring.
Hiring Update: Communications Coordinator on the Horizon
The Digital Communications Coordinator position is progressing through the hiring process. This role will support a range of digital initiatives, from training staff on Microsoft applications to coordinating with the Digital Innovation Task Force on bigger-picture strategy.
Kelly Adkins, the Health Department’s new IT, digital systems, and communications specialist, is already working on a comprehensive training plan for Health Department staff on Microsoft apps, launching in January. This kind of systematic approach to digital literacy is exactly what the Communications Coordinator will help scale across the county.
There’s a real need here: lots of people know how to log into Teams or Zoom, but fewer understand the embedded tools that make virtual collaboration actually effective. Breakout rooms, chat moderation, screen sharing with specific windows – these features can transform meetings from “I guess we’re all on video now” to genuinely productive collaboration.
Action Items
- AI User Group Survey – All task force members to share the survey link with their professional organizations
- Virtual Meeting Infrastructure – Explore hearing room D virtual setup options; share Health Department’s Microsoft training communications plan; identify volunteers to help organize the first joint meeting
- FOIA Form Router – Finalize routing questions and implement on live site
- Skills Testing – Complete full workflow test and report results
- St. Joseph Coordination – Schedule and coordinate first joint virtual meeting
The government AI coalition summit delivered one clear message that echoed throughout today’s meeting: data quality, not AI quality, is the bottleneck. Counties across the country are discovering that their AI initiatives fail not because the technology isn’t good enough, but because the information the AI is drawing from is incomplete, inconsistent, or inaccessible.
Which means the unglamorous work of data cleanup, standardization, and organization isn’t just busywork – it’s the foundation for everything we’re trying to build. Good news: we’re already working on it. Better news: we’re not the only ones who’ve figured this out.
See you at the next meeting, possibly in a virtual room somewhere, possibly with St. Joseph County joining us. Democracy adapts.
