Every Circle Square client has an engineer whose name is on their account. Every AI agent we build should have one too.
At Circle Square, every client has a primary engineer who owns the account. When something breaks, the client isn’t sending a ticket into the void. They know whose name is on their environment, and that engineer knows it too.
When we started Signet, our AI division, I wanted the same thing for every agent we build. A name on it.
Where the habit comes from
I spent six years as a KC-10 crew chief in the Air Force, out of McGuire. The rule on the flight line was simple. Every aircraft had an owner. Every job was signed for. Nothing left until the checklist was done. Check everything, assume nothing.
My signature on a form meant I had done the work and I was willing to be asked about it. Nobody wanted to be the name on a write-up that came back wrong, so we worked hard to make sure it didn’t.
It’s an old idea, and not only a military one. A single-barrel bourbon label tells you which barrel it came from. That isn’t decoration. It’s there so someone can trace it back.
The question that usually gets a pause
Now look at how a lot of businesses are adopting AI. Someone buys a tool. Someone connects an agent to the inbox, the ticket queue or the shared drive. It drafts emails, summarizes contracts, sorts requests. Much of the output is good, and the people who set it up had good intentions.
Then I ask one question. When this gets something wrong, who is responsible?
Most of the time there’s a pause. Sometimes the answer is the software company. Sometimes it’s whoever clicked install. Rarely is it a person who reviewed how the agent behaves, signed off on it and expects to be asked about it.
That’s fine for a birthday message. It’s harder when the work is going to a client, an auditor or a board. Those readers want to know who stands behind it.
This isn’t anyone’s fault. AI moved fast, and in most places the accountability part hasn’t caught up yet. That’s the part we decided to start with.
How Signet signs the work
Our line at Signet is “AI does the typing. A named human makes the judgment calls.”
Every agent we deploy runs inside what we call the Harness. It starts with a context pack: your rules, your systems, your obligations, your house style. The agent drafts. A named reviewer checks the work against a written checklist. Then we sign it, with logging and an escalation path. Low-stakes work, like sorting and tagging, can run on its own. Nothing high-stakes ships unsigned.
So every agent is reviewed, logged and owned by a named engineer. When someone asks who decided what and on what basis, the answer is already on record. Nobody has to rebuild it from memory or go digging through a support portal.
We use the same approach on our own internal work before we bring it to clients. If it isn’t good enough for us to sign, it isn’t ready for yours.
What a signature actually does
Most days, nobody asks to see the log. That’s fine. The signature still does its job, because the engineer who owns the agent knows the question could come. So they set it up properly, they review what needs reviewing and they keep shortcuts from piling up.
It’s the same reason our clients have a primary engineer and not a rotating queue. Ownership changes how people work. It did on the flight line, and it does with software.
If you’re thinking through where AI fits in your business, or you already have a few tools running and aren’t sure who owns them, I’m happy to talk it through. Ready now or just curious, either one is a good place to start.
One standard. Every client. The door is always open.
If you want to see how an agent gets reviewed, logged and signed before it touches your business, start with how the Harness works. circlesquareconsulting.com/signet/how-it-works