Most of what's written about AI and construction estimating is either hype or fear. The real, current picture is duller than both: it's useful for the repetitive first draft of a quote, and it still needs a human to check it before it goes anywhere near a client.
01 — Photo and plan takeoffs
Apps that estimate rough material quantities from site photos or PDF plans are genuinely useful for a first pass — fast, and they catch the obvious. They're not a substitute for walking the site, and they don't know about the thing behind the wall the plan doesn't show.
02 — Drafting the written quote
Turning site-visit notes or a voice memo into a formatted quote document saves real admin time — often the most tedious part of estimating. The wording still needs a human read-through before it goes to a client; a confidently-worded mistake is still a mistake.
03 — Price benchmarking
Flagging when a materials line looks unusually high or low against recent jobs is a useful sense-check, not a source of truth. Treat it the way you'd treat a second opinion — worth having, not worth following blindly.
04 — What it's not good at yet
Judging site-specific risk — access, existing condition, the things likely to go wrong that a plan won't show — is still a judgement call built on experience, not something a model can price from a photo. That's exactly the kind of thing this issue's Cornerstone piece is about.
Used right, it buys back an hour on the admin side of a quote. Used blindly, it puts a confident-sounding number in front of a client that nobody's actually stood behind. Treat the output as a draft, not a quote.