Why roofing work stays on the roof
Roofing is decided inch by inch, in weather, on a slope. A crew tears off old shingles, checks the deck for soft spots, lays underlayment, then nails courses so every overlap sheds water the right way. Cementing and nailing flashing around chimneys, vents and valleys is where leaks start, and no two roofs present the same angles. That is hand work with judgment attached.
The setup is physical too. Ladders, hoists and scaffolding go up before anything else, materials get moved to the work face, and the crew reads the pitch, the wind and the heat as they go. Software can plan a job, but someone still has to stand on it. Our scoring treats that share of task time as the core of the answer, and you can read how the three questions are weighed on the methodology page.
There is also a second check on timing: demand. The Bureau of Labor Statistics counts about 135,490 roofers in the United States, with median pay of $55,440 and projected employment growth of 5.3% from 2025 to 2035 (BLS, 2025). Office work in roofing companies is changing faster than field work.
What AI does, helps with, and leaves to people
The tasks our data puts fully with software are the paper ones: measuring roof area and slope from aerial imagery, and turning those measurements into a materials list and a written scope. That group covers 6% of task time. It is the part of the job a homeowner never sees.
Assisted tasks sit in the middle, at 0% of task time. Inspecting a problem roof to work out a repair is one: drone photos and image models flag missing tabs, blistering or ponding, then a roofer confirms the cause on the deck. Scheduling crews and ordering materials is another, where software drafts and a person decides.
Everything else stays with people, and that is 94% of task time: laying and fastening shingles or panels, sealing flashing at joints and penetrations, and rigging access before work starts. The overall coverage figure, our answer to can AI do it, is 4 out of 100, and the way that number is built is set out under how coverage is measured.
How strong the evidence is
Weak, and we say so plainly. The evidence grade for our parity question, is it better than a person, is D. That grade means no study has tested an AI or robotic system against a working roofer on roofing tasks, so we publish no parity number for this job. Benchmarks that score office and coding work do not reach a tear-off.
What would settle it is specific. A timed trial on a pitched residential roof, with a machine or AI-directed system doing tear-off, underlayment, fastening and flashing, judged by a journeyman roofer and then by leak and warranty outcomes a year later. Published results on fastening accuracy and valley detail would move the grade. Until then the score leans on task structure, not on a head-to-head test. You can see how the open dataset handles jobs with thin evidence across the full job rankings.
When the picture could change
Most likely after 2048 (8 in 10 of our scenarios). Two things would pull that earlier. One is hardware: almost all of this job is physical, and the robot class our data says it would take is a dexterous humanoid, so cheaper and more reliable machines in that class change the math. The other is prefabrication, where panelized roof sections are built in a factory and craned into place, moving hand work indoors to machines that already exist.
Two things push it later. Conditions: steep pitches, wind, heat, hidden rot and roofs that were framed wrong fifty years ago defeat fixed routines. And accountability: permits, code inspections, manufacturer warranties and insurance claims all expect a licensed person to sign for the work. What the dated range does and does not claim is explained on the replacement-year method page.
What to do: get fluent with the measurement and imagery tools your estimator uses, so the office work moving to software moves with you rather than past you.
How roofers stay needed
Lean into the tasks that stay on the roof. Diagnosing a leak from the deck up, detailing flashing at chimneys, valleys and vents, and rigging safe access on an awkward building are all work a model can describe but not do. Crews that own those three are hard to thin out.
Two skills compound. First, reading a roof’s history: decking, ventilation, ice damming, prior repairs, and what a manufacturer’s spec requires for the warranty to hold. Second, leading and estimating, including checking what the measurement software returns before it becomes a bid. Entry-level hiring is where AI shows up first in this trade, because the paperwork a new hire used to learn on is the part software now drafts.
Nearby trades share the same shape of work and the same exposure. Compare the detail on helpers, roofers, sheet metal workers and insulation workers, or look at the wider picture for construction trades workers and the construction sector.
Two more steps are worth a few minutes. Put roofing beside another trade you are weighing on the side-by-side comparison tool, and see which hands-on jobs cluster together on the list of jobs that mostly need a person.