Jobs / Architecture and engineering
Can agents do architecture and engineering work?
Architects, drafters, and engineers across civil, mechanical, electrical, and more.
- US jobs
- 2,617,130 (1.7% of employment)
- Typical wage
- $101K a year (employment-weighted mean of occupation medians)
- Occupations
- 59 in O*NET 31.0
- Evidence
- 2 direct benchmarks, 2 also relevant, 3 context only
The answer today
Some tasksFirst measured since 10 Oct 2025.
Rule: Agents complete 20% or more of the tasks (or projects) on a direct benchmark.
The evidence that sets this step
- HWE-Bench by Peking University: 79.9% of tasks completed · GPT-5.6 Sol max (Codex CLI), 31 Jul 2026
Direct evidence that disagrees
- Parametric CAD Bench V3 by gNucleus AI: 9% of tasks fully completed (Perfect tasks) · Claude Opus 5.5 (max) with Claude Code 2.1.280, 24 Sep 2026
Next step: Most tasks needs 60% or more of the tasks (or projects) on two direct benchmarks from independent institutions.
How much of the work is measured
3% of the work in this family is measured by direct benchmarks. Little of the work is measured (less than 20%). On 10 Oct 2025: 0%.
Work = the O*NET 31.0 work activities of the family's occupations, weighted by how important each task is and by US employment. A benchmark measures an activity when its tasks make the agent do it and its grading checks it. Method.
| Work activity (O*NET) | Share of the work | Direct benchmark |
|---|---|---|
| Create visual designs or displays | 6.5% | Measured |
| Evaluate designs, specifications, or other technical data | 5.3% | Measured |
| Document technical designs, procedures, or activities | 3.3% | Not measured |
| Research technology designs or applications | 3.0% | Not measured |
| Design structures or facilities | 2.9% | Not measured |
| Advise others on business or operational matters | 2.5% | Not measured |
| Design electrical or electronic systems or equipment | 2.5% | Measured |
| Communicate with others about specifications or project details | 2.2% | Not measured |
| Determine resource needs of projects or operations | 2.1% | Not measured |
| Direct organizational operations, activities, or procedures | 2.1% | Not measured |
Our reading
Agents complete some architecture and engineering tasks. On Peking University's HWE-Bench, GPT-5.6 Sol with Codex CLI resolves 79.9% of 417 hardware bug-repair tasks in open-source chip projects (one run per row, all by the maintainers). Parametric CAD Bench V3 disagrees: the best agent scores 61.03% on 100 FreeCAD tasks, but that is partial credit, and no row gets full credit on more than 9 tasks. Both test narrow slices: chip repair and CAD parts. No direct benchmark measures documenting technical designs or designing structures. The Remote Labor Index has CAD and architecture projects, but it is broad and sets no step. A second independent benchmark at 60% or more moves the family up.
Evidence: HWE-Bench, Parametric CAD Bench V3, Remote Labor Index by CAIS, gNucleus AI, Peking University, and Scale AI. Written 10 Oct 2026.
Benchmarks for this work
Direct evidence
Work benchmarks mapped primary to this family. They set the step when they report how many tasks agents complete.
- 61%Claude Opus 5.5 (max) with Claude Code 2.1.280 · 24 Sep 2026Strong
Also relevant
These benchmarks cover part of this family's work, or have no result yet. They do not set its step.
- 64.8%Claude Code + Opus 5.5 (max) · 6 Oct 2026Strong
Context only
These benchmarks span many occupations, or they test how agents behave. They give context but do not set the step.
Functions: Professional deliverablesHardware design and debuggingIssue resolutionComputer useEnterprise workflowsCAD modelingFreelance projectsTerminal operationsFeature developmentML engineering
What we do not know
- Benchmarks measure tasks, not whole jobs. A job is a bundle of tasks, judgment, and relationships.
- Capability is not adoption. A task that agents can do in a benchmark may still be done by people at work.
- Most results come from single runs. Reliability over many runs is often unknown.
Occupations in this family
Sorted by US employment (BLS OEWS May 2025). O*NET sub-occupations share the employment figure of their SOC group. Work measured = the share of the occupation's O*NET work that direct benchmarks test.
How to cite
Credit the original work first: the benchmarks listed on this page, and their institutions.
Then, if you used this page:
Can Agents Work. "Can agents do architecture and engineering work?." https://canagentswork.com/jobs/architecture-engineering/ (accessed 2026-10-11). CC BY 4.0.@misc{caw-jobs-architecture-engineering,
title = {{Can agents do architecture and engineering work?}},
author = {{Can Agents Work}},
year = {2026},
howpublished = {\url{https://canagentswork.com/jobs/architecture-engineering/}},
note = {Accessed 2026-10-11. CC BY 4.0}
}