What AI changes in the job, task by task
The most exposed tasks are directing testing and validation (a score of 0.65 out of 1) and analysing the requirements of an application (0.61). Designing, writing and fixing the software itself sits at the threshold (0.50). Tasks built on conversation are less exposed: evaluating hardware and software interfaces with engineers (0.45) and discussing maintenance with customers (0.45).
Developing and directing software testing and validation procedures
ExposedResearching, analysing and evaluating requirements for software applications and operating systems
ExposedAssessing, developing, upgrading and documenting maintenance procedures for operating systems, communications environments and applications software
ExposedDirecting software programming and development of documentation
ExposedResearching, designing, and developing computer software systems
ExposedModifying existing software to correct errors, to adapt it to new hardware or to upgrade interfaces and improve performance
ExposedConsulting with engineering staff to evaluate interfaces between hardware and software
LowConsulting with customers concerning maintenance of software system
Low
Source: ILO, Generative AI and Jobs, Working Paper 140, 2025.
Neighbouring jobs
In IT, nearly every occupation is exposed, to different degrees. Web development sits at the highest level of the index; applications programming at the same level as software development (gradient 3). Systems analysis, technical support and IT management are less exposed (gradient 2).
Web and Multimedia Developers
Very highly exposed
Applications Programmers
Highly exposed
Software Developers· this job
Highly exposed
Systems Analysts
Moderately exposed
Information and Communications Technology User Support Technicians
Moderately exposed
Information and Communications Technology Service Managers
Moderately exposed
Limits
The ILO index is global and describes the tasks of the international classification, not a specific position: "developer", "software engineer" and "programmer" all fall under code 2512. Its 2025 scores were largely produced by AI models, then checked against a survey of workers and expert panels: they are estimates, not measurements. Opening counts only cover Brevy's catalogue.