■ MODERATE RISK ■ Science & Research
Software will take over more of the modeling and report-writing that fills geotech billable hours, but the ground keeps being weird and someone licensed has to stake their stamp on what's under the building. Expect fewer hours per project, not fewer stamps.
“AI models soil behavior. Your core samples are just confirmation now.”
Our AI replacement risk score — how we score jobs
Geotechnical engineers answer one deceptively simple question — will the ground hold this up? — through site investigations (drilling boreholes, logging soil, running CPTs), lab testing, and analysis: bearing capacity, settlement predictions, slope stability, seismic response, retaining-wall and foundation design. The deliverable is usually a report a structural engineer and a contractor will rely on, sealed by a licensed professional who becomes legally liable if the parking garage settles differently than promised. Fieldwork is muddy; the analysis end is spreadsheets, FEM models, and specialist software like PLAXIS.
Automation is compressing the analysis middle. Machine-learning models trained on regional borehole databases now predict soil profiles between borings, AI-assisted interpretation reads CPT data instantly, and generative tools draft the boilerplate-heavy geotech report that once consumed junior-engineer weeks. Parametric design software iterates foundation options automatically, and sensor-laden instrumentation streams settlement and pore-pressure data that algorithms monitor better than monthly site visits. The pattern: fieldwork stays physical, but the interpret-model-report pipeline needs fewer human hours every year.
What resists is uncertainty and liability. Soil is nature's least standardized material — models trained on regional data still get ambushed by an unmapped peat lens or uncharted fill — and knowing when the model's tidy answer smells wrong is exactly the judgment that separates engineers from software. Construction disputes, contaminated sites, and karst terrain demand experience. Above all, the professional-engineer stamp is a legal institution: someone insurable must certify the foundation design, and no jurisdiction is licensing an algorithm. Our risk score of 45 fits a profession whose grunt work automates while its judgment and liability core stays scarce — infrastructure spending and climate-driven ground problems are even growing it.
Automatability: our editorial assessment of current and near-term AI capability
AI-assisted interpretation and report drafting are entering firms now, and by the early 2030s the analysis-and-writing hours per project will have shrunk substantially — juniors feel this first. Licensed judgment, fieldwork oversight, and expert consultation stay human through 2040 and beyond, propped up by liability law and growing infrastructure and climate-adaptation demand. The profession transforms into a leaner, more supervisory shape rather than contracting outright.
The routine analysis and reporting layers are — software increasingly interprets soundings, predicts profiles, and drafts reports. The profession itself is protected by two stubborn facts: soil surprises models regularly, and the law requires a licensed human to seal foundation designs and absorb the liability. Add rising infrastructure and climate-adaptation work, and demand for judgment looks solid even as hours per project fall.
Yes — it's chronically under-chosen compared to structural, which keeps salaries and demand healthy, and the field's judgment-heavy nature ages well against automation. Just plan for a career where AI handles the parametric grinding and your value is field insight, licensure, and communication. The engineers who struggle will be those whose only skill was running analyses software now runs itself.
Mostly as acceleration: machine-learning interpolation of subsurface conditions between borings, automated CPT and lab-data interpretation, monitoring platforms that watch instrumentation continuously, and drafting assistance for reports. Firms treat outputs as first drafts requiring engineer review, because training data rarely covers the site's specific geology. It's a productivity story so far — with clear implications for junior headcount.
Field judgment that catches what models miss, fluency with data-driven characterization tools, and the consultative skills for claims, expert testimony, and client advisory work. Specialties tied to spending trends — offshore wind foundations, levee and slope resilience, urban retrofit — offer the strongest demand. Licensure remains the non-negotiable foundation; everything automatable orbits around the engineer who signs.