EXPOSED
The modal biological technician runs defined protocols at a bench — sample prep, pipetting, cell culture, PCR, running instruments, logging results — plus a meaningful share of screen work (data tabulation, QC checks, drafting run summaries) that AI already handles. The hands-on portion is real protection, but it happens in a controlled lab, which is exactly where liquid handlers, automated colony pickers, and integrated plate-reader pipelines are already displacing manual steps. There's no license, no signature requirement, and little client relationship, so nothing outside the tasks themselves shields the role.
Mixed — a routine tier and a judgment tier. Plating cells, maintaining cultures, feeding animals, calibrating a microscope and troubleshooting a clogged HPLC line still require hands at the bench, but the other half of the day — Excel tabulation, plate-reader output cleanup, ELISA curve fitting, drafting the section of the methods writeup the PI edits — is already being done by software, which is what pins this at 10 rather than the 15+ of a field ecologist.
Some physical or field component. The work is physical but the environment is engineered to be predictable — a BSC hood, a fixed 96-well footprint, an incubator at 37°C — which is precisely the geometry Hamilton and Opentrons liquid handlers were built for; the 12 reflects the genuinely awkward residue (dissecting tissue, coaxing a temperamental centrifuge, animal handling), not the routine pipetting.
No licence, no signature requirement. Nothing you produce carries your name in a legally meaningful way: the PI signs the grant report, the study director signs the GLP protocol, the CLIA lab director signs out clinical results, and your role in the audit trail is an initial in a notebook margin that the 3 accounts for.
Anonymous artifact production. Your outputs move to a PI, a study director or a QC reviewer who cares about the CV of the triplicates, not who ran them; the 4 rather than 0 is the institutional knowledge of which freezer holds which lot and which incubator drifts — real, but it transfers with a two-week handover.
Executes defined procedures on defined inputs. You work from SOPs, IACUC-approved protocols and validated assay procedures where deviation requires a supervisor's sign-off, so the discretion is narrow — deciding a plate is contaminated and rerunning it, or flagging that a standard curve failed acceptance criteria — which is a 6, not the 10+ of someone choosing the experimental design.
Has AI actually changed your work?