I fixed a production bug by voice at a gas station
I caught a scheduling bug in our rafting operation.
A van and driver had been assigned correctly, but between sync issues, scheduling conflicts, and last-minute booking changes, there were no longer enough seats for everyone on the return shuttle. Not a major operational problem. We caught it early.
A few weeks earlier, this would have cost me about an hour: tracing the scheduling logic, finding the bug, working out why the conflict engine hadn’t fired, updating the code, testing it, deploying a fix. Realistically it would have waited until evening, because the middle of an operating day is not when you sit down to debug.
Instead I pulled into a gas station.
While the tank filled, I opened a Claude Code session on my phone and described the problem out loud. It searched the codebase, found the bug, updated the scheduling logic, restored the missing conflict checks, committed the changes, and had the fix staged with a single confirmation left before deployment.
By the time I was back on the road, it was ready.
The same workflow now handles most of what used to be a queue: debugging, scheduling assignments, working out operational constraints, optimizing how we use vehicles and crew. All of it through ordinary conversation.
The speed is the part that’s hard to internalize
Problems that used to sit in a backlog for days, or eat an afternoon, now get identified, fixed, and deployed in the time it takes to fill a gas tank.
That changes how you think about building software. It changes how you run a business a lot more.
When a fix costs an hour and a context switch, you batch problems. You let small things accumulate because each one isn’t worth stopping for. The backlog becomes a permanent feature of the operation, and everyone quietly learns to work around the things in it.
When a fix costs four minutes at a gas station, that logic inverts. Small problems get fixed the day they appear, which means they never compound, and the operation stops carrying a standing tax of known-but-unaddressed issues.
What it doesn’t change
I’m still the one deciding. The system found the bug and wrote the fix, but the judgment about whether that fix was correct, and whether shipping it mid-season was wise, stayed with me. Anything touching money, compliance, or safety still requires a human confirmation and reports back exactly what changed.
That constraint isn’t a limitation. It’s the reason the crew trusts the system enough to use it.