Note · AI Implementation Strategies

Stop Waiting for the GreenLight. Build the Alarm Instead.

David He, FounderAugust 28, 20263 min read

Nobody is coming to say AI is allowed. What owners actually want is to find out their automation broke before a customer does.

Nobody is coming to tell you AI is allowed. What most business owners actually want is smaller than permission anyway: they want to know that if they build the thing, they will find out it broke before a customer does.

A room that had already settled the big question

I spent a Thursday afternoon at an Ask Us Anything About AI forum in Greenville. Thirty small business owners in an old Pepsi bottling plant, three panelists, a hundred minutes that ran long. Plumbing, handyman work, financial planning.

In all that time, nobody asked whether AI was worth using. They had settled that on their own.

What they asked, over and over, was what the rules are. Can I put a client contract in there. What do I tell customers about their data. How do I let my team build things without it getting away from me.

That is not fear of the technology. That is a room of people who know what they want to build, standing at the edge of it, waiting for someone qualified to say go.

One owner said some of his employees are further along with AI than he is. He did not want to stop them. He wanted someone with a title to tell him it was allowed.

The question I keep coming back to

A woman near the front asked it. What catches a scheduled automation that quietly stops running?

Not one that errors. One that simply does not fire on Tuesday, and you notice in October.

The panel said have a person verify it. That works, and it costs you a person.

The cheap version takes fifteen minutes

Your automation pings a URL every time it finishes. If that ping does not arrive on schedule, you get a text.

It is called a heartbeat check, or a dead man's switch. Healthchecks.io and Cronitor do it free at small volume.

It alerts on silence, which is the point. A job that never ran throws no error to catch.

Keep a register, not an approval gate

For the owner who wanted permission: do not build an approval gate. Keep a register.

One shared sheet. One row per thing anyone builds, what it touches, and who to call when it stops.

Gates get routed around. Registers get filled in, because a row takes thirty seconds and protects the builder.

The one place AI made more work

I asked a question too. I am a software engineer and my job is making things safe to put in front of real customers, so I asked whether AI had made more work than it saved.

It has, in one place. Producing got nearly free. Checking did not.

Scale up what you generate without touching how you check it and you have not saved time. You have moved the pile.

If you were not in that room

You have a version of this too. Something you already decided was a good idea, sitting still because nobody official has said it is fine.

Stop waiting for the green light. Build the thing, then build the alarm that tells you when it breaks.

More notesnewest first

Working on something like this?

Bring the app or the process to a free 15-minute call. I will tell you what I would look at first, and whether I am the right person for it.