Teaching the system to catch its own mistake
Running architecture in house meant we had to build our own second opinion, so I added an adversarial review layer.
With architecture in house, catching our own errors before they turned into costly field change orders fell to our team, with no outside firm's second pair of eyes to rely on.
I put an adversarial review layer inside the in-house design process. Its specialist agents existed to hunt for contradictions in the model, not just to help produce it faster.
Speed with no check is how a quick in-house process becomes a costly field mistake. If you remove the outside firm's second opinion, you have to rebuild your own on purpose, structured so it cannot simply agree with the first pass.
On its own, the review system flagged a bay that was nine and a half feet off and a wrong setback, worth roughly $250,000 in avoided rework, before any person found either one.
A human still decides whether to trust or override what the system flags. The tool found the error, but a person had to act on it. The system serves better judgment and does not replace it.
Story details
| Project | 301 W Osborn |
|---|---|
| Type | Teaching moment |
| Year | 2026 |
| Firm telling | on the Common Ground wiki |