Protect the work practitioners love.
Agentic Colleague is our investigation into AI that can take on bounded AEC work while project evidence, professional judgment, and the authority to proceed remain with the practitioner.
The parts of AEC practice worth protecting are not only the final decisions. They include the careful reading, the small acts of coordination, and the moments when experience changes what a team chooses to do next.
Repeated work can be assisted without pretending that judgment is routine. The boundary matters: the task can move, but authorship and responsibility do not disappear with it.
One conversation, accountable work.
Project work rarely arrives as an isolated prompt. A question carries earlier decisions, constraints, drawings, messages, and the person who must review what follows. We are exploring one continuing project conversation where each request stays bounded to the work at hand.
A proposed next step should return with the material that shaped it. The practitioner can then inspect the work, correct it, or decide that it should not move forward.
Learning alongside a practice.
Every practice has its own language, standards, preferences, and ways of checking work. Useful assistance has to learn from the context a practice deliberately brings into a task, not flatten that context into a generic answer.
That learning should remain scoped and inspectable. It supports the practice rather than claiming ownership of its knowledge or quietly changing the terms of the work.
Built to show its working.
The request, the evidence, the assumptions, and the proposed next step should remain available for review.
This is the standard guiding the work. A colleague earns trust by making the path to a proposal clearer, and by stopping where practitioner authority begins.