Traffic control, not the pilot

If you run a small firm, you have probably had the same year we did. AI was supposed to be the staff you couldn’t hire. Instead it invented a decision nobody made, rewrote a design intent into something plausible and wrong, and forgot on Tuesday what it agreed to on Monday. It was fluent and confident, and it was no help at all.

We didn’t quit. We changed its job.

The frustration was a job description problem

We had been asking AI to decide: draft the answer, pick the option, fill the gap. That is exactly where it fails. When a model doesn’t know, it doesn’t say so. It writes something that sounds right. In a design practice that’s dangerous, because a plausible wrong answer about a door, a circuit or a scope line costs real money once it reaches the field.

So we stopped asking it to fly the plane.

What air traffic control actually does

A controller doesn’t choose where the plane goes, and doesn’t touch the controls. A controller knows where every aircraft is, keeps them apart, puts them in order, and says something the moment two paths are about to cross. The pilot decides. The controller makes sure the pilot is never deciding blind.

That is the job we gave AI. Three things:

• Capture. What was decided in the meeting becomes a record that day, with what it changes and what it replaces.

• Route. Each decision goes to the drawing, the schedule or the person it affects.

• Flag. When something has no origin, no owner, or contradicts an earlier decision, it shows up, in red.

What it doesn’t do: choose a finish, approve a substitution, talk to a contractor, or set a fee. Those are ours.

What it showed us

The first time we drew our own record as a map, it was humbling. We had 86 recorded decisions and not one line connecting any of them. The studio’s own system was documented more carefully than the design work it was meant to serve. On a completed project, we had delivered construction support the contract didn’t cover, and nobody had flagged it, including us.

None of that was a technology failure. It was what happens in every small firm: decisions get made in the room and live in someone’s head. Traffic control is what made it visible.

The first real thread we traced was a door. In June we chose a custom millwork door. In July, in the field, it became a flat solid door. The June decision existed only in a meeting transcript. Once it was recorded and linked, anyone could see what the field had changed and why. That is the whole idea in one door.

One rule that keeps it honest

AI proposes. A person confirms. A connection between two decisions only exists if the newer one names the older one, or someone who was there says “same door.” A good guess stays a guess. That rule is the difference between a tool that helps and a tool that fills your record with confident fiction.

What this means for you as a client

• What you decide in a meeting is written down that day, not reconstructed from memory weeks later.

• When something changes on site, you can see what it replaced and who agreed.

• Scope is tracked down to the meeting where it was asked for, so there’s no surprise at the end.

• Your designer spends the time designing, not chasing what was said.

• The judgment stays human. The design decisions are ours and yours.

Questions we get

Does AI design my space?

No. It keeps the record of how your space was decided. The design is ours, made with you.

Does AI talk to my contractor or vendors?

No. Anything that leaves the studio is reviewed and sent by a person.

Then what’s the point?

Nothing you say gets lost, and everything that changes can be traced back to why.

Read the long version

We wrote up the method, the findings and what we haven’t solved yet as a working paper for other small-practice owners: What your project record is not telling you.

Amaze. Elevate the spirit. Have a positive impact.

Start a conversation →