Dustin Schuler clocks in around 7 a.m. Pacific, but his work started the night before. The marketing director at Bud Clary Auto Group, a 14-store group based in Kelso, Washington, leaves several AI sessions running overnight in Claude, ChatGPT or whichever model fits the job. When he logs on, most of the work is done, and his job is to check it.
"I'm just... kind of human in the loop, just correcting what we built overnight," Schuler told host Daniel Kim on AI for Dealers.
His reasoning is simple. Sitting and waiting on AI responses during business hours can eat a whole day, so the long jobs run after hours.
What runs overnight at a 14-store group?
Website work. Schuler used Claude Code to rework the group's Dealer Inspire site toward an ecommerce feel, then trimmed bloat so pages load and buttons respond faster.
QA across every site. Alerts flag him when an OEM push or a change breaks something on any store's website.
Reports built from several systems. One target was a report that pulls from three systems and would save a service director five hours a week.
Scheduled scripts. Cron jobs build BDC lists and send emails every day without anyone asking.
How does he decide what's worth automating?
"A lot of things are important, but not everything's urgent," Schuler said. His filter is whether a project saves time or money, and how much.
He runs napkin math on vendor tools, too. A $900 monthly website widget, at an average $30 per lead, has to produce 30 leads. At a 10% close rate, that's three cars a month just to break even. Multiply it across 14 rooftops and replacing it in-house starts to look obvious. His team has already brought several website widgets in-house.
What's the catch?
Everything AI builds overnight still has to be supported during the day. "I also have to support it when the tickets come in, when things break," Schuler said. Security matters too, especially after a run of PHP exploits in October.
That's why Bud Clary expects to have two people working full time on AI development by next month, and why Schuler wants developers with computer-science backgrounds. Someone who can code uses AI to go faster. Someone who can only prompt is guessing at what's under the hood.
Where should you start?
Pick one workflow and explain it to AI the way you'd train a new employee:
What you do, step by step, and how often.
What you've already tried, and what didn't work.
Where you get stuck.
Talk it out if typing slows you down. Schuler's advice is to start small enough to prove the concept, then hand it something big.
Pick the task you dread most each week, explain it to AI tonight like you're training a new hire, and review what it hands back in the morning.
If AI worked your store's night shift, what's the first job you'd hand it, and who would check its work in the morning?

