EXPOSED
The core of this job is physically handling instruments — mounting a pressure transducer on a deadweight tester, aligning optical comparators, torquing fixtures, running a thermocouple through a bath — which today's robotics cannot do across the messy variety of customer equipment and plant floors. What AI does erode is the paperwork half: calibration certificates, uncertainty budgets, drift trend analysis, procedure write-ups, and ISO 17025 documentation packages are all text-and-spreadsheet work that automates well. Accreditation regimes require a named, competent technician to sign results, which slows substitution without legally locking it the way an engineering stamp or nursing license does.
Mixed — a routine tier and a judgment tier. Automated calibration software already sequences multimeter and pressure-controller runs, captures readings over GPIB/Ethernet, and generates as-found/as-left tables without a human touching a keyboard — but someone still has to select the standard with adequate TUR, build the fixture, decide whether a failing gauge gets adjusted or condemned, and troubleshoot an instrument that drifts nonlinearly, which is why this sits at 12 and not down in the automatable band.
Hands-on in uncontrolled environments. Work happens on plant floors, in temperature-controlled labs, and at customer sites: hauling standards in a van, mounting torque wrenches and transducers, running thermocouples through salt and oil baths, cleaning and shimming optical comparators, working around live 480V panels and pressurized lines — 15 rather than 18 only because a meaningful share of the day is spent in a bench-and-terminal environment you control.
Certification preferred, not legally required. ISO/IEC 17025 and ANSI/NCSL Z540.3 require a named, technically competent person to authorize each certificate, and ASQ CCT or NCSLI credentials are routinely demanded by accredited labs — but that is an accreditation-body finding against your employer, not a state licence that can be revoked from you personally, so it delays substitution rather than blocking it.
Some relationship component. Repeat aerospace and pharma customers ask for the tech who knows their specific gauge population and audit history, and being the person the quality manager calls before an FDA or AS9100 audit has real value — but the deliverable is a certificate traceable to NIST, and a different accredited lab issuing the same numbers is an acceptable substitute, which caps this at 7.
Meaningful discretion. Most decisions run off written procedures and published tolerances, but you own the genuinely ambiguous calls: whether an out-of-tolerance finding triggers a reverse-traceability notification on every part measured since the last cal, whether to guard-band a marginal pass, and how to justify an uncertainty budget contribution an assessor will challenge — discretion inside a documented framework, hence 9.
Has AI actually changed your work?