Agents in the Trades
Quality

Punch-list photo chase

Closeout on a commercial solar array or a tenant buildout comes down to the punch list. Every item has an owner and a due date. Every item also needs a photo showing the work is done: the torque...

The problem

Closeout on a commercial solar array or a tenant buildout comes down to the punch list. Every item has an owner and a due date. Every item also needs a photo showing the work is done: the torque marks on the lugs, the relabeled disconnect, the patched drywall, the re-sealed roof penetration.

The work usually gets done. The proof doesn't. A foreman fixes the item, gets pulled to the next problem, and never sends the photo. The superintendent ends up walking the site with the list, texting owners one at a time, and asking, "Did you close out number 14?" Then they scroll through a group chat looking for a picture that may or may not exist.

That is photo police work. It eats time near the end of a job when time is shortest, and it delays sign-off because an item without proof can't be closed.

How the agent workflow works

The agent watches two things: the punch list and the place photos land. The list can be a spreadsheet or a punch tool. The photo location can be a shared folder, a project app, or a dedicated email or text number that crews send to.

Each morning, the agent reads the list and finds every item that is past due or due within a set window, such as two days, and has no photo attached or matched.

For each one, it drafts a short reminder to the item's owner. The message names the item, the location, the due date, and exactly what the photo should show, pulled from the item description. For example: "Item 14, Inverter 3 DC disconnect label. Due Thursday. Send a photo showing the new label installed and legible."

When a photo arrives, the agent reads the filename, the message text, and the image itself. It matches the photo to an open item and drafts a note: "Photo received for item 14. Appears to show a labeled disconnect. Ready for your review." It does not close the item.

The superintendent has two checkpoints. First, they approve the morning batch of reminders before anything goes out, at least until they trust the wording. Second, they close items themselves. The agent only moves an item to "proof received, needs review." The superintendent looks at the photo and decides whether it shows finished work.

At the end of the day, the agent sends one summary: items closed, items awaiting review, items still missing proof, and owners who have been reminded more than twice.

What you need to set it up

Plan on a few hours to clean up the list format and write the description field properly. Plan on another hour or two to test with a handful of real items. The description field matters most, because the agent can only ask for the right photo if the list says what the right photo is.

Where it breaks

Photo matching is the weak point. If a crew sends a picture with no item number and no caption, the agent may guess wrong or leave it unmatched. A close-up of a label looks like any other label.

The agent can't judge workmanship. It can tell a photo shows a disconnect with a label. It can't tell whether the label is the right one, or whether the lug was torqued to spec. That is why the superintendent closes items.

Bad list data produces bad reminders. Vague descriptions like "fix as needed" lead to vague requests. A wrong owner means the reminder goes to someone who can't act on it.

Reminders can also become noise. If the agent pings a sub every morning about the same item, people tune it out, especially when the real problem is that the sub is waiting on material. The agent doesn't know that unless someone writes it on the list.

Finally, spotty cell coverage on remote solar sites means photos arrive late, in batches, or after the daily run.

The bottom line

This workflow doesn't fix slow crews. It takes the repetitive chasing off the superintendent and leaves the judgment where it belongs. The agent finds the gaps and drafts the nudges. The superintendent approves the messages and decides what counts as done. Start with the list and the photo habit. If those are clean, the agent is useful within a week of closeout.

Related use cases

← All use cases

One working agent workflow, every day.

Real AI agent use cases from solar, construction, and field operations — what they do, what they cost, and where they break. No hype, no theory.