Where AI actually helps in estimating (and where it doesn't)

Most 'AI estimating' marketing is software trying to look modern. Here's where AI earns its place in a real estimator's workflow — and where it gets in the way.

We built Bedrock with AI baked in. We also believe most "AI estimating" marketing is asking software to do work it can't do. Here's where we drew the lines.

Assembly authoring. Describe a 12" stem wall with #5 horizontal at 16" OC and #4 vertical at 24" OC. The AI returns a ready-to-edit assembly with the formulas, parameters, and outputs. You verify and save. This is genuine 10× faster than building it by hand. Plan metadata extraction. Sheet numbers, drawing types, scale annotations — all extractable with high accuracy. Bedrock OCRs every sheet on upload so you don't manually rename 300 sheets. In-app help. Asking "how do I calibrate scale on a sheet that doesn't have a scale bar?" should not require a support ticket. Our in-app assistant answers it from product docs.

Auto-takeoff. "Upload a plan, AI traces every footing." Marketing claim. Reality: 60–80% accuracy on the easy stuff, 0% on the hard stuff. An estimator has to validate every measurement anyway. If you have to validate, you might as well have drawn it. We don't ship this. We won't. AI margin recommendations. Your margin is a strategic decision based on your relationship, the GC, the labor market, and what you're willing to walk away from. Software that recommends a margin number is software that doesn't understand the bid. Auto-bid generation. No.

Plan comparison. Diffing addendum revisions against a base set is genuinely hard for humans and reasonably tractable for AI. We're working on this. Quantity sanity checks. "Your slab CY is 12% lower than typical for this SF." Useful as a flag. Dangerous as a decision.

Machine-readable site summary for AI crawlers: /llms.txt