Software developers, quality assurance analysts, and testers

Document applications and systems for future maintenance and upgrades

AI can generate comprehensive documentation from code, including API references, architecture descriptions, and maintenance guides. Tools already auto-generate docs from comments and code structure with minimal human input, requiring only light review for completeness.

BLS evidence: Developers 'Document every aspect of an application or system as a reference for future maintenance and upgrades.'

Write computer code for software applications or systems

Current AI code generation (GitHub Copilot, GPT-4, Claude) can write substantial portions of application code from specifications, including boilerplate, standard algorithms, and common patterns. Developers increasingly review and integrate AI-generated code rather than writing from scratch, matching the anchor example.

BLS evidence: The page states 'some developers write code themselves instead of giving instructions to programmers.'

Create models and diagrams to guide programmers in coding

AI can generate UML diagrams, flowcharts, and architectural models from natural language descriptions or existing code. Tools like Mermaid integration in LLMs produce diagrams automatically. Humans review for accuracy but the generation itself is largely automated.

BLS evidence: Developers 'Create a variety of models and diagrams showing programmers the software code needed for an application.'

Create test plans and execute software testing procedures

AI can generate test cases from requirements, write unit and integration tests, and execute automated testing. Tools already create test plans and identify edge cases. Human oversight remains for test strategy and interpreting complex failures, but labor content has shrunk significantly.

BLS evidence: Quality assurance analysts and testers 'Create test plans, scenarios, and procedures for new software' and 'Implement software testing, using either manual or automated programs.'

Identify and document software defects and usability problems

AI-powered testing tools can identify bugs, performance issues, and usability problems through automated analysis and pattern recognition. Documentation of defects can be auto-generated from logs and stack traces. Humans validate severity and prioritization.

BLS evidence: Quality assurance workers 'Document and report defects or problems with software' and 'document and track the software's potential defects or risks.'

Design application architecture and plan how components will work together

AI systems can now propose system architectures, generate component diagrams, and suggest design patterns based on requirements. Tools like GPT-4 with code understanding can draft architectural decisions, though humans review for organizational fit and non-functional requirements.

BLS evidence: Developers 'Design each piece of an application or system and plan how the pieces will work together.'

Perform software maintenance and ensure programs continue functioning normally

AI can identify deprecated dependencies, suggest patches, refactor code for compatibility, and automate routine maintenance tasks. However, understanding legacy system context and making judgment calls about breaking changes requires human oversight, placing this in the batch-review category.

BLS evidence: Developers 'Ensure that a program continues to function normally through software maintenance and testing' and 'may perform upgrades and maintenance' after release.

Identify project risks and recommend mitigation steps

AI can analyze project data, code complexity, and historical patterns to identify technical risks and suggest mitigations. Risk assessment tools increasingly use ML to flag schedule, security, and technical risks. Humans review recommendations but AI does substantial analytical work.

BLS evidence: Quality assurance analysts and testers 'Identify project risks and recommend steps to minimize those risks.'

Recommend software upgrades for existing programs and systems

AI can analyze codebases to identify technical debt, security vulnerabilities, and opportunities for modernization, then recommend specific upgrades. However, evaluating business impact, migration risk, and organizational readiness requires human judgment that keeps humans central to the decision.

BLS evidence: Software developers 'Recommend software upgrades for customers' existing programs and systems.'

Analyze users' needs and design software to meet those needs

AI can generate requirements documents and suggest architectures from user interviews, but translating ambiguous stakeholder needs into concrete software designs requires iterative human judgment and negotiation that AI cannot fully replace. AI assists substantially but humans remain load-bearing.

BLS evidence: Software developers typically 'Analyze users' needs and then design and develop software to meet those needs' and 'may begin by asking how the customer plans to use the software so that they can identify the core functionality the user needs.'

Provide feedback to developers and stakeholders on usability and functionality

While AI can analyze usage patterns and generate usability reports, providing contextual feedback to stakeholders requires understanding team dynamics, political considerations, and translating technical issues into business language. AI assists with data gathering but humans deliver the feedback.

BLS evidence: Quality assurance analysts and testers 'Provide feedback to software developers and stakeholders regarding usability and functionality.'