How we use AI to run a creative agency without letting it touch the creative.

Most of the AI advice aimed at agencies is about making content with it. We went the other way. Clients hire us because a real person walked into their restaurant, noticed the light hitting the bar at 4pm and shot it. If a machine does that part, we have nothing left to sell.

So we drew a line. AI runs the business, and people make the work. We automate everything on the business side as hard as we can. Nothing on the creative side gets touched.

1. We built our own software instead of renting five apps

A typical agency stack is a scheduler, a project board, an approval app, a file link and a reporting dashboard. Then someone spends hours a week copying information between them.

We built one client portal instead, with Claude Code as our developer. Clients log in to see their content calendar, approve posts, leave comments and read their reports. It syncs both ways with the Notion boards we work from, so a comment left in either place shows up in both.

A couple of years ago, custom software was out of reach for a two-person shop. Now the hard part is deciding what you want it to do. Start with one form or one dashboard, or whatever you copy and paste every week.

2. Automations that remind, and people who send

The portal runs a few quiet jobs in the background. One nudges us when a post has waited on a client too long, so we can follow up before the posting date slips. Another sorts client comments by what they need, so a change request never gets buried under a thumbs-up. A third drafts follow-ups for new leads in my voice and puts them in a queue.

The rule that keeps this safe: the automation prepares and a person decides. Onboarding, contracts and billing never go out unless Mary or I click the button ourselves.

3. Backups we can actually trust

We shoot a lot of video, and every shoot lands on an external hard drive first. Drives fail. So every shoot gets backed up to Google Drive, filed by client and by shoot, the same way every time.

AI runs the whole job. It scans the drive, compares it with what's already in the cloud, uploads what's missing, retries when the internet drops and keeps a log.

We learned one lesson the hard way. A "done" message is not proof. Early on we were told everything was backed up, and it wasn't. Now every backup gets checked file by file, by exact size, against what's in Drive. The first full audit found 930 client videos that had never made it up. All of them are backed up and verified now. Trust AI, then check its work the way you'd check a new hire's.

4. Finding the right clip in seconds

After a few years of shooting, storage stops being the problem. Finding things is. "Do we have a golden-hour rooftop cocktail?" used to mean an hour of scrolling through folders.

Now we keep a map of where every finished photo and video lives, by client and by industry. Our Lightroom catalog has more than 17,000 photos, and we can search it by idea, like "plated seafood from above" or "pour shot at night." Pulling references for a new client's mood board takes a couple of minutes.

5. Prospecting with a person in the loop

Cold outreach is where AI writing gets lazy fast, and people can tell. Our setup researches each business first: their story, their recent posts, what makes them different. Then it drafts an email or DM in my voice. I read every one, and nothing goes out until I approve the batch.

AI does the homework a good assistant would do. The relationship is still mine.

6. Reports and reminders

A monthly client report used to eat an afternoon. Now the numbers come straight from Instagram and Facebook into a branded report with a draft email. We read it, add what we noticed from the work itself, and send it.

The rest is small stuff that adds up. Shoot reminders, approvals due, invoices, renewals, the backup queue. Two people don't have a project manager, so the reminders do that job.

What we'll never hand to AI

The shoot happens on location, and so does reading the room. The edit is taste: color, pacing, which frame leads. Clients hear from Mary or me, never from a bot. And AI can suggest a caption, but we decide what goes out under a client's name.

How to start using AI in your agency

1. List what you repeat every week. Backups, reminders, reports and follow-ups are a good place to look.
2. Write down what AI will never touch. Ours is anything creative, plus anything a client receives without a person reading it first.
3. Automate one thing and get it trusted before you add the next.
4. Keep a person approving anything that sends, spends or deletes.
5. Spot-check every new automation until it earns your trust.

Curious about the creative side? Here's how we shoot cocktail bar photography, the part of the job we keep completely human.

FAQ