EXPOSED
This is a catch-all bucket — biosecurity analysts, ecological researchers, lab-based bioinformatics and regulatory science staff — and the modal worker splits time between bench or field protocols and screen work: literature synthesis, dataset cleaning, statistical runs, manuscript and grant drafting. The screen half is exactly what LLMs and AutoML now do at usable quality, which compresses the analyst tier hard. What persists is hands-on assay execution, field sampling in messy conditions, and deciding which experiment is actually worth running and whether a result is real.
Mixed — a routine tier and a judgment tier. Sequence QC, differential-expression pipelines, literature triage for a review, and first-draft methods sections are already handed to tooling in most labs in this bucket, but designing an assay that discriminates between two competing mechanisms, troubleshooting why the western blot failed on the third replicate, and running a field transect still take a person — a roughly even split, which is why this sits at 10 rather than in the resistant band.
Some physical or field component. Bench work here is pipetting, culture maintenance, dissection, and instrument runs in a controlled BSL-1/2 lab, with periodic field sampling — soil cores, water grabs, trapping — that is genuinely uncontrolled but seasonal rather than daily, so it lands above pure desk science and below the year-round outdoor roles that score 15+.
No licence, no signature requirement. No state licence gates the title: you can run this work with a PhD or a master's and nobody countersigns your results, and the constraints that do exist — IRB and IACUC approval, select-agent registration, IBC review — attach to the institution and the protocol, not to your name, so the 4 reflects committee oversight rather than personal legal exposure.
Some relationship component. Your outputs are peer-reviewed papers, agency reports, and internal data packages that are judged on the numbers rather than on who produced them, though sustained relationships with a program officer, a collaborating field agency, or a long-running study site do carry weight that a stranger cannot instantly replicate — enough for 6, not the named-advisor dependence of clinical or consulting roles.
Meaningful discretion. You decide when a surprising result is an artefact versus a finding, whether the effect size justifies another year of sampling, and how to word an uncertainty caveat that a regulator or a biosecurity customer will act on — real discretion with real consequences, but almost always filtered through PIs, co-authors, and reviewers before it binds anyone, which caps it at 12.
Has AI actually changed your work?