The client
Poma Architectural Metals (pomametals.com) designs and manufactures architectural metal — railings, cladding, ornamental work — for commercial buildings across the U.S. Every project is custom, and every project starts with a drawing.
Their engineering team is talented but small. As order volume grew, the bottleneck became clear: senior engineers were spending hours per drawing on work that didn't require senior judgment — measuring, extracting cut dimensions, building BOMs, and preparing shop-ready outputs.
The bottleneck — in their words
When UTS first sat in on a working session, the math was stark. A typical custom railing drawing took an engineer 90 to 180 minutes from received-from-architect to ready-for-shop. Of that, roughly 60% was deterministic mechanical work that an AI agent could do: parsing the AutoCAD file, extracting member dimensions, generating cut lists, and producing the output package.
The remaining 40% required the engineer's judgment — verifying load assumptions, resolving ambiguous customer specs, choosing the right finish process. That part stayed with humans.
The build
UTS shipped a custom AI agent that integrates directly with Poma's AutoCAD environment. The agent ingests new drawings as they arrive, identifies repeating parts, generates parametric cut lists, validates against Poma's standard part library, and routes the output package to production.
Critically, every output is reviewable. The engineer sees what the AI did, with confidence scoring on each step, and approves or corrects before anything hits the shop. The agent learns from corrections, so accuracy compounds over time.
- Drawing ingest: AutoCAD file in, structured parts catalog out
- Parametric cut list generation with material optimization
- Standard-part library validation against Poma's catalog
- BOM and shop package output, formatted for the floor
- Engineer review interface with confidence scoring and one-click corrections
Results
After 6 weeks of build and 4 weeks of tuning in live production, the system replaced ~5× the throughput of the previous manual workflow. Engineers shifted from drafting to design judgment — the work that actually requires their experience.
Equally important: the system runs every day, processes real customer work, and is part of Poma's standard production flow. It's not a pilot. It's infrastructure.
What the operations lead said
"UTS didn't come in with a pitch deck. They came in, learned our process, and in two weeks we had a working AI system on our floor. That's the difference."
Why this generalizes
Most manufacturers have a version of this problem: highly experienced engineers spending too many hours on the deterministic part of a drawing workflow. The Poma pattern — AI handles the mechanical work, engineers keep the judgment work, every output is reviewable — generalizes to fabrication, prefab, structural, and architectural metalwork shops.
If your engineers feel underutilized doing repetitive drafting, the AI build is straightforward. UTS will scope a free proof of concept on one of your real drawings.