The rule: AI doesn't sign
In an industry where an engineer's signature is personal liability, the first question about any new tool is not how clever it is but who is answerable for its output. Our answer is fixed: AI prepares, filters and surfaces — a person decides and signs. That rule is precisely what lets us put the technology into real workflows instead of leaving it in demos.
Where it already works
KORA, the system we built at BIM TEAM to run our work, is where this happens. Inside it, AI handles the work nobody wants to do by hand:
- Automatic classification of incoming documents and routing them to the right project
- Connections between the systems we work with, so a figure is recorded once
- Real-time dashboards on project status instead of a report assembled by hand at month end
- Automation of repetitive production and management processes
What actually changed
The meaningful change was not the speed of any single task. It was how long it takes for a problem to become visible. When data is recorded in one place and displayed continuously, a deviation shows up while there is still time to act on it. That is why we invested in an internal system rather than settling for external tools.
Where it still doesn't work
AI does not design structures. It does not replace engineering judgment, it does not carry responsibility for code compliance, and it does not remove the need for human review of anything heading to construction. Models that produce convincing text also produce convincing mistakes, and in engineering the cost of a convincing mistake is far higher than the cost of being slow.
How we decide where AI goes
- Is the task repetitive and well defined?
- Is there a way to check the output before it affects anything?
- Does a human remain accountable for the decision?
Three yeses and the tool goes into the process. One no and it stays out.