Going further 4 min read
How do we turn an AI idea into a real workflow?
Write the job as steps and watch the handoffs. A locate request, worked department by department.
Part of the path We already use AI
Write the job down as steps before you ask AI to do any of it, and pay most attention to the handoffs. A workflow is the same job done the same way every time: the same inputs, the same order, the same review points. In local government, the work usually stalls in the handoff between departments, so that’s where AI and a person’s check belong.
Where cross-department work actually stalls
Most local government jobs that feel slow are slow because the work waits: in an inbox, for a reply, for someone to notice it arrived. The planner finished her part on Tuesday. Engineering saw it Friday.
That’s why a good workflow is written around handoffs. For each step, ask three questions: what comes in, what goes out, and who gets it next. AI can take the reading and writing at each step. A person checks before the work moves on.
A locate request, department by department
One customer asked how to take a single idea, a locate request, and turn it into a project. A locate request arrives when a contractor plans to dig, through Ontario One Call in Ontario or the 811 system in the US. The town has a few days to confirm whether its water, wastewater, or other buried assets are in the way. The request passes through several teams, and every handoff is an email.
| Step | Today | Where AI takes the first pass | Who checks |
|---|---|---|---|
| 1. The request arrives | Someone reads the ticket and retypes the address, dig area, dates, and work type into an email. | Reads the ticket and pulls those details into a standard summary. | The coordinator confirms the address and dates. |
| 2. Mapping checks the location | Mapping looks up the site and writes back which assets are nearby. | Drafts the note to water and wastewater from mapping’s findings, in the same format every time. | Mapping confirms what it found. |
| 3. Water and wastewater respond | Each team replies separately, in its own words. | Combines both replies into one status: clear, mark, or needs a site visit. | The utilities lead signs off. |
| 4. Public works acts | Public works schedules a crew to mark the site, or doesn’t. | Drafts the crew note with the site details already filled in. | The public works supervisor schedules it. |
| 5. The reply goes out | Someone writes the clear or mark response to the contractor. | Drafts the response from the combined status. | The coordinator sends it. |
Notice what didn’t change. Mapping still checks the map. Utilities still decide. Public works still marks the site. What changed is the retyping and rewording between them, which is most of the waiting.
Write your own workflow on one page
Pick the job you chose as your first one. For each step, fill in:
- Who does it: the role as well as the department
- What comes in: the document, form, or email
- What goes out: the finished piece, in the format the next person needs
- Where it waits today: be honest; this is where the time goes
- What AI drafts: the reading, sorting, or writing
- Who checks before it moves on: a named person
If you can’t fill in a step, that’s useful too. It usually means nobody owns that handoff today.
Make it repeatable
A one-off request can be typed fresh each time. A workflow can’t. When the same job comes in every week, give AI the same instructions every time: which of your own documents to use, the shape the answer should take, and when to hand the question to a person. The answer then comes back in the same structure, with the source it drew on, so staff can check it quickly.
That’s also when a workflow becomes an Agent: a job written down well enough that AI can take each step across teams, with the review points built in.
A shorter example: a council briefing
The same approach works inside one department. Take a council briefing on proposed downtown parking changes:
- Summarize the background material.
- List the likely impacts on residents, businesses, and city operations, including risks.
- Draft the briefing with set sections: background, key impacts, recommended next steps.
- Review it for tone and clarity for a public audience.
A person reads each stage before the next one starts. That’s where errors get caught early, and where the job stays yours.
What to watch for
- Don’t automate a broken process. If two departments disagree about who does a step, settle that first.
- Keep the decision with a person. AI can say “this looks clear.” A person decides whether it is.
- Start with one path. Handle the common case first and send the unusual ones to a person.
Most workflows depend on your own documents: the bylaw, the standard, the template. The next question is how to make AI use them: How do we get AI to answer from our own policies?