EXPOSED
The core production work — digitizing features from aerial and satellite imagery, generating contours and DEMs from stereo pairs, symbolizing and labeling map sheets, georeferencing and mosaicking scans — is exactly what machine-learning feature extraction and automated cartographic generalization now do at usable quality and enormous scale. What holds is control-network and accuracy work: setting ground control, validating positional accuracy against standards, resolving conflicting source data, and signing off on deliverables that carry legal weight for boundaries, floodplains, or utility corridors. In several states photogrammetric mapping tied to property or engineering use must be performed under a licensed surveyor, which is a real but state-patchy shield.
Core tasks are already automatable. Stereo compilation, feature digitizing, contour derivation, edge-matching and label placement are all now closed loops in ArcGIS Pro, SOCET GXP, Pix4D and Agisoft — you press a button on the point cloud and get breaklines and DEMs that used to take you a week, which is why this sits at 6 rather than mid-band; only ground-control planning and accuracy QC survive as genuinely manual.
Some physical or field component. Most of the week is a dual-monitor stereo workstation, but field visits for control-point recovery, photo-ID of panel points, checking that a bridge deck or tree canopy in the imagery matches the ground, and occasional flight-planning or drone operation put a real physical tail on the job — enough for 6, not the 12 of someone who is outdoors weekly.
Certification preferred, not legally required. An 8 reflects the state patchwork: in Texas, Florida and others photogrammetric mapping for property or engineering purposes must be sealed by a licensed surveyor, but a GIS-track cartographer producing thematic or planning maps needs no licence at all, and CP/GISP certification carries no personal legal exposure.
Anonymous artifact production. Deliverables are orthomosaics, DEMs and map sheets that get consumed by an engineer or planner who will never learn your name — the metadata and accuracy report do the talking, so at 5 there is no client relationship to lose to a vendor pipeline.
Meaningful discretion. You decide which of two conflicting parcel sources controls, whether a checkpoint residual is a blunder or terrain, and whether a deliverable meets ASPRS or NSSDA tolerances — real calls, but bounded by published standards and reviewed upstream, which is why it lands at 9 rather than in the ownership band.
Has AI actually changed your work?