EXPOSED
The core of this job is physical: dragging and coupling loading arms and hoses, climbing tank cars to open hatches, setting valves, grounding and bonding, watching gauges, and pulling samples — none of which today's robotics does at terminal scale. The real threat isn't AI reasoning but terminal automation that has been eating this occupation for decades: automated loading racks with card-reader driver kiosks, mass-flow meters, overfill sensors, and remote console control let one operator supervise many bays, and AI-assisted monitoring pushes that ratio further. Paperwork tasks — bills of lading, shipping papers, batch records, tank gauging math — are already largely software-handled.
Mixed — a routine tier and a judgment tier. At 13 rather than 17, the hose-and-hatch work genuinely resists software, but you already work next to racks where a driver swipes a card, keys a product code, and the preset meter shuts off on its own — the loading itself is a solved control problem, and what's left is connect, inspect, and troubleshoot; that's what keeps it out of the 6-and-below band but off the top.
Hands-on in uncontrolled environments. 17 reflects climbing to the dome of a car in the rain to open a hatch, torquing a stuck valve, dragging a 4-inch vapor hose into position, and doing it in a Class I Division 1 area where you're bonded and grounded — uncontrolled outdoor environments with confined-space and fall exposure, only short of 20 because the work happens on a fixed rack you know inch by inch rather than at unpredictable sites.
Certification preferred, not legally required. 10 sits at the top of the certification band because HM-126 hazmat function-specific training under 49 CFR 172.704 and the CDL/tanker-endorsement world genuinely gate who touches the product, but the certificate is not a personal licence — when a car overfills or a placard is wrong, DOT and PHMSA cite the shipper of record and the terminal operator, not you.
Anonymous artifact production. 4 because the truck arrives, the seal number matches, and the driver leaves; the customer downstream reads a meter ticket and a certificate of analysis, and no one at the receiving terminal knows or asks which loader made the connection.
Meaningful discretion. 9 recognises the real calls you make — refusing a car with a bad valve or missing gasket, stopping a load on a pressure anomaly, deciding a sample looks off-spec — but those decisions run against written loading procedures, product-compatibility charts, and a supervisor you can raise on the radio, so the ambiguity is bounded rather than yours to own alone.
Has AI actually changed your work?