AI for Metal Fabrication: What Works Today

What are fabricators and glass shops using AI for now?

Mostly office tasks, and adoption is early. The best trade data covers glaziers: in Glass Magazine's 2026 Top 50 survey, published June 17, 2026, about a quarter of respondents were in the early stages of using AI, over a third actively used it, and only 10% had implemented it company-wide. Most said they use it in preconstruction, specifically for estimating. Other uses were contract review, turning staff knowledge into procedures, and meeting notes. These are large firms reporting on themselves.

No comparable 2026 figure for metal fabricators was found. Several blogs quote adoption percentages with no named survey, and those are left out here. This table sums up the rest of the post, with our own read of the evidence:

Can AI handle quoting intake?

Yes, for sorting the package before an estimator starts. Paperless Parts says its AI reviews RFQ emails and files for part information, builds line items and assigns files to them. Its 2024 press release headline claims 90% less quote setup time. That is the vendor's number, and the release gives no method. It was built for job shops and contract manufacturers, so a Division 05 bid package is a different test.

Reading prints for risk is newer. In September 2025 the same company announced Requirements Review, in private beta, to pull tolerances, notes and specs from prints. Its CEO named the real danger: when a shop misses a requirement, it risks underquoting, and underquoted jobs are the ones you win most often. A tool like this is a second reader. It does not take the responsibility off your estimator.

AI reads a clean package better than a messy one, so fix the front door first with an RFQ page that collects the right details.

Does AI takeoff work for steel and glass?

For structural steel, it can give a fast first pass. Steel Genie, sold by SDS2 and ALLPLAN, detects beams, columns, joists and other members across a drawing set and builds an estimating model. The vendor says one customer, Master Steel, raised estimating throughput 67%. That is one shop, reported by the seller. ALLPLAN's own case study would not open. The same page says accuracy depends on the quality and clarity of the drawings, and it builds in a review step, which is the honest part. Ferra, another steel tool, describes itself as estimator-in-the-loop, with estimators responsible for scope.

Miscellaneous metals is less settled. Steel Genie's page lists capabilities for stairs and rails, but it gives no accuracy data for them. For storefront and curtain wall, vendor pages exist, and a search for an independent test of any glazing takeoff tool found none.

You can run the test yourself. Take three finished jobs, one clean and two messy, run them through a trial, and compare against your real takeoff and final quantities. Count what it missed, since an error that shrinks the bid costs more than one that grows it.

Can a shop trust ChatGPT with drawings and pricing?

Use general chatbots for drafts, and keep them away from pricing and customer files. NIST describes confabulation as generative AI confidently presenting false content, and calls it a natural result of how these models work. A wrong weld size written with confidence is worse than a blank.

Customer drawings are the second risk. Read your NDAs before you paste anything into a public tool. If you handle defense work, ITAR registration rules are only the start, so ask your attorney. As for pricing, a 2024 piece in The Fabricator expects AI to inform pricing but concedes its predictions aren't 100% accurate. Nothing opened for this post shows AI pricing custom work well without a person.

Can AI do follow-up for a fab or glass shop?

Yes, for first replies to web forms and messages, and this is the lowest-risk place to start. It can ask a homeowner for opening sizes, glass type and photos, or ask a buyer for quantity and date. The best-known data is old and indirect: the 2007 InsideSales.com and MIT study found the odds of contacting a web lead dropped 100 times between 5 and 30 minutes. It measured phone calls to web leads, was published by a vendor, and did not address close ratios.

Phone calls are a separate problem, and missed call text back has its own rules. Anything with a bid date, such as an invite that came through your GC bid list work, goes to a person.

Where should a shop start with AI?

Start where a mistake is cheap and you can check the output:

• Web form replies and detail collection.
• Meeting notes, procedures and first drafts of emails, read before sending.
• RFQ sorting, with an estimator checking each line.
• Takeoff trials on finished jobs, measured against your actuals.
• Pricing last, if at all. GHOBZ's AI agents for glass and metal shops cover the first step, and the overview for metal and glass businesses shows the rest of the work.