软件开发人员
计算机与数学AI暴露度
- 数据来源: BLS发布时间: '26.08
很高· 相对
低四级相对区间很高职业群单位数值
尺度 · 母数 · 来源
四级相对类别（低 / 中等 / 高 / 很高）
以BLS就业预测表中831个细分职业为基数。取值按 NEM（全国就业矩阵）代码给定，因此共用同一 NEM 代码的职业得到相同分档
- 数据来源: Anthropic发布时间: '26.03
0.288
0.000此处所载数值的范围0.745尺度 · 母数 · 来源
- 数据来源: ILO发布时间: '25
0.53
0.09此处所载数值的范围0.70职业群单位数值
尺度 · 母数 · 来源
生成式AI暴露度指数，按发布原值0–1
ISCO-08职业小类单位 — 共用同一代码的职业取值相同
本站计算，并非国际劳工组织发布的数值。在本站与 ILO 数据集相连的 1,012 个职业中，达到或高于此值的占 14%。
该来源发布的是什么性质的数值
BLS的类别是相对排位而非绝对水平，也不是一手测量：它把多项已发表研究给出的职业百分位排名归并为四个等级。它不是就业或工资预测，不是采用概率，也不区分自动化与增强。
Task-level exposure
显示隐藏的 11 项工作| Task | Claude.aiRaw / share % | APIRaw / share % |
|---|---|---|
Modify existing software to correct errors, allow it to adapt to new hardware, or to improve its performance.15-1132 | 5.974659.1 | 9.960757.1 |
Modify existing software to correct errors, to adapt it to new hardware, or to upgrade interfaces and improve performance.15-1133 | 2.566625.4 | 4.597226.3 |
Analyze user needs and software requirements to determine feasibility of design within time and cost constraints.15-1132 | 0.46444.6 | 1.46648.4 |
Design, develop and modify software systems, using scientific analysis and mathematical models to predict and measure outcome and consequences of design.15-1132 | 0.26692.6 | 0.21851.3 |
Store, retrieve, and manipulate data for analysis of system capabilities and requirements.15-1133 | 0.21512.1 | 0.50062.9 |
Analyze information to determine, recommend, and plan computer specifications and layouts, and peripheral equipment modifications.15-1132 | 0.11881.2 | 0.02810.2 |
Develop and direct software system testing and validation procedures, programming, and documentation.15-1132 | 0.08410.8 | 0.32071.8 |
Evaluate factors such as reporting formats required, cost constraints, or need for security restrictions to determine hardware configuration.15-1133 | 0.07380.7 | 0.00560.0 |
Direct software programming and development of documentation.15-1133 | 0.07100.7 | 0.05470.3 |
Consult with customers about software system design and maintenance.15-1132 | 0.05970.6 | 0.00840.0 |
Values in this tab are predicted labels, not observations. Eloundou et al. (2023) published two rating regimes — human raters and GPT-4 — and the β shown here is derived from the GPT-4 rater basis alone; the same task can take a different value under the other regime. The unit and the meaning differ from the observed shares (%) in the other tabs, so do not place them on the same axis.
| Task | βE1 + 0.5 × E2 |
|---|---|
Analyze information to determine, recommend, and plan installation of a new system or modification of an existing system.O*NET Task ID 21661 | 1.0 |
Analyze user needs and software requirements to determine feasibility of design within time and cost constraints.O*NET Task ID 21662 | 1.0 |
Confer with systems analysts, engineers, programmers and others to design systems and to obtain information on project limitations and capabilities, performance requirements and interfaces.O*NET Task ID 21664 | 1.0 |
Coordinate installation of software system.O*NET Task ID 21666 | 1.0 |
Design, develop and modify software systems, using scientific analysis and mathematical models to predict and measure outcomes and consequences of design.O*NET Task ID 21667 | 1.0 |
Determine system performance standards.O*NET Task ID 21668 | 1.0 |
Develop or direct software system testing or validation procedures, programming, or documentation.O*NET Task ID 21669 | 1.0 |
Modify existing software to correct errors, adapt it to new hardware, or upgrade interfaces and improve performance.O*NET Task ID 21670 | 1.0 |
Monitor functioning of equipment to ensure system operates in conformance with specifications.O*NET Task ID 21671 | 1.0 |
Obtain and evaluate information on factors such as reporting formats required, costs, or security needs to determine hardware configuration.O*NET Task ID 21672 | 1.0 |
β = E1 + 0.5 × E2 · E1 = tasks where direct LLM access alone cuts time by at least 50%, E2 = tasks where software built on top of an LLM cuts time by at least 50%. Values take only 0 / 0.5 / 1.0.
Data sources & licenses — O*NET®, Anthropic Economic Index, Eloundou et al. (2023): see full notices on the Credits page
Occupation information
与本职业相关的近期变化
2026年7月: ADP/Stanford linked postings-payroll study of ~7,000 IT workers (2019-2025) prices tasks separately within IT jobs. Advising others on the design or use of technologies is among 8 higher-wage activities; five tasks lost compensation value in 2023-2025 vs 2019-2022, including "develop models of systems, processes, or products." Effect sizes were not published.
[来源: ADP Research / Stanford Digital Economy Lab, Unbundling Jobs (July 2026)]2026年6月: Cited by ADP Research as an example of a high-AI-exposure occupation. Group-level payroll data shows employment in high-exposure occupations down 0.2% year over year overall and down 4.3% for workers aged 22-25 (33rd consecutive monthly decline). The percentages are for the high-exposure group, not for this occupation alone.
[来源: ADP Research, Canaries Dashboard (June 2026)]2026年4月: Korean youth employment data shows codified knowledge work most vulnerable to AI displacement. Professional/technical services lost 98,000 workers in Jan 2026 (worst since 2013). Entry-level coding tasks at highest risk.
[来源: Econmingle / National Assembly Budget Office (2026)]