At the federal level, most things are big, underfunded, understaffed, underpaid, and over regulated, so programs are low quality, behind, and slow. I stay away from state level, and expect there is less data and $, and even more problems.
Think fairly basic data extraction, data cleaning, workflow automation, moving to the cloud, getting basic modern security.
It may help to think about baselines. The status quo is being late or bad, understaffed, low budget, etc. It is not comparing to perfect code written by top CS grads with all the time in the world.
Developers need help writing basic code, and for fancier tasks, help writing models better than their bad rules & bad ML.
Also, data is frustrating in these contexts. For example, consider data cleaning, where AI is a pretty clear win for spotting bad input errors, performing entity resolution, extraction of dirty data, etc. Federal level is by definition pretty big, and data comes from all sorts of terrible systems, often staples together from a hodgepodge of them. The shift to ML/AI here has been going for 10+ years, eg, Trifacta.
A little more out there, a lot of forms and data sources dealing with people is text, not just nicely structured data, so being able to more easily work with that is going to keep a lot of people busy for awhile.
Related to the last point, and I stay away from this side, but gov is also ultimately very much a service industry. Automating citizen interactions, and cheapening the cost to do it well, is a big deal.
Think fairly basic data extraction, data cleaning, workflow automation, moving to the cloud, getting basic modern security.