Rules first. AI for judgment.

The fastest way to lose trust in AI is to use it for work that rules already do well.

Most of what slows down engineering delivery is not a judgment problem. Reinforcement schedules, sheet set-ups, model clean-ups, parameter checks and quantity take-offs are repeatable. They follow rules, and rules can be automated in a way that is predictable, auditable and cheap to run. Putting a language model in front of that work makes it slower, less consistent and harder to check.

AI belongs where judgment helps: interpreting a design question, finding the right clause in a standard, proposing options for a clash, or explaining why two models disagree. Even there, the engineer stays in control and signs off.

Four principles for how this scales

• Rules first. Rule-based automation does the repeatable work.
• Everything connected. Every tool talks to every other tool, with no manual hand-offs.
• AI, adopted wisely. Used where judgment helps, with engineers in control.
• MCP and web apps. MCP connects agents to proven tools; web apps put them in reach of every team.

Rules for the repeatable. AI for judgment. Adopt wisely. Connect everything.

The result is not an AI strategy and an automation strategy running side by side. It is one system, where agents call the same proven tools engineers already trust.