Generative AI for Construction: Specs, Estimates, and Jobsite Knowledge

An estimator uses a commercial AI tool to research concrete pricing trends for a bid. The AI generates a confident answer with plausible numbers. The estimator uses them in the estimate. Three months later, when the concrete is actually purchased, the prices are 18% higher than what the AI suggested — because the AI was drawing from training data that was 12 months old and didn't account for the regional market conditions in the project's geography. The bid margin, which was 4%, is gone. The estimator's judgment call would have been more accurate — he would have called two suppliers and gotten current pricing in 20 minutes. The AI saved 15 minutes and cost the company the job margin.

Generative AI that works in construction requires grounded retrieval against current, authoritative sources. The project specification (not a generic specification). The company's historical cost database (not internet averages). The current supplier pricing (not training data from last year). OSHA standards (not paraphrased safety advice). Contract terms (not generic legal language). With these grounded, the AI produces useful, sourced answers. Without them, it produces confident wrong answers that construction professionals can't afford to rely on.