Computer Programmers
15-1251.00Median wage $100,390/yr92,230 employed (US)Rank #42 of 923 scored · top 5% by substitution
Create, modify, and test the code and scripts that allow computer applications to run. Work from specifications drawn up by software and web developers or other individuals. May develop and write computer programs to store, locate, and retrieve specific documents, data, and information.
Sub-scores
0–100 · band = confidence interval from rater disagreement
Substitution59
Exposure51
Augmentation87
Substitution — the headline: capability discounted by cost, barriers and adoption.
Exposure — technical capability alone, regardless of whether anyone deploys it.
Augmentation — how much AI assists without replacing. High here + moderate substitution = a changing job, not a disappearing one.
Tasks on the substitution scale
17 rated tasks, binned by substitution score.
Position among all scored occupations
Distribution of 923 occupation scores; the marker is this occupation.
Tasks with substitution ≥ 70
35%
Run 1.0.0-draft.1 · computed 2026-08-05 · rater panel: claude-sonnet-5, claude-haiku-4-5-20251001 · intervals span rater disagreement.
Why this score
The five weighted dimensions of the composite, averaged across this occupation's tasks (importance-weighted, panel mean). Exact weights and formulas: /api/v1/methodology.
Task automatabilityw 35%50
panel mean rating 3.0/5 → substitution pressure 50/100
Technical feasibility todayw 20%54
panel mean rating 3.1/5 → substitution pressure 54/100
Cost vs. human wagew 15%61
panel mean rating 3.4/5 → substitution pressure 61/100
Adoption barriersw 20%inverted — strong barriers lower the score71
panel mean rating 2.1/5 (barrier strength) → substitution pressure 71/100
Sector adoption velocityw 10%74
panel mean rating 4.0/5 → substitution pressure 74/100
Task breakdown (17 tasks)
Substitution pressure per task, weighted by O*NET importance in the composite. Expand a task for the full rater audit trail — every rating, every model, every rationale.
Conduct trial runs of programs and software applications to be sure they will produce the desired information and that the instructions are correct.
84CI 77–91 · exposure 80 · augmentation 88 · importance 4.0/5 · click for rater detail
Conduct trial runs of programs and software applications to be sure they will produce the desired information and that the instructions are correct.
84| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 5/5 | Continuous integration and automated testing are standard practice across software development, from startups to enterprises. Adoption of AI-assisted test generation and debugging is accelerating rapidly in high-digitization, fast-moving tech sectors. |
| Sector adoption velocity | claude-sonnet-5 | 5/5 | Software engineering is among the fastest and deepest AI-adopting fields, with automated testing, CI pipelines, and AI coding agents already embedded in production workflows. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI assists programmers by running tests, analyzing failures, suggesting fixes, and generating test cases, significantly raising debugging productivity. Humans remain in the loop for strategy and judgment, but AI transforms the speed and thoroughness of trial runs. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI dramatically speeds up writing test cases, running trial executions, and diagnosing failures, while the programmer still directs the process and interprets results. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Trial runs and testing can be substantially automated using test frameworks, continuous integration/deployment systems, and AI-assisted test case generation. AI can execute, compare outputs, and flag deviations from expected results, reducing manual testing time by >50% in many scenarios, though edge cases and novel failure modes may still require human oversight. |
| Task automatability | claude-sonnet-5 | 4/5 | AI coding assistants and agentic tools can write, run, and interpret test cases, execute programs, and flag discrepancies with substantial time savings, though complex integration testing still needs human judgment. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Few regulatory or legal barriers exist to automating trial runs; organizations can freely choose AI-assisted testing. Some friction remains from legacy systems and organizational inertia, but nothing structurally prevents adoption in most contexts. |
| Adoption barriers | claude-sonnet-5 | 1/5 | No licensing or regulatory requirement mandates a human to conduct trial runs; this is a routine technical QA task with no legal sign-off requirement. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | Automated testing via CI/CD and AI tools costs a fraction of manual testing labor per execution cycle. Once infrastructure is set up, each trial run is nearly free at scale, making the cost multiple orders of magnitude lower than paying a human tester for equivalent coverage. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Running automated test suites and AI-assisted debugging is far cheaper per iteration than manual trial runs by a programmer, though oversight and occasional human re-verification add some cost. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 5/5 | Mature, production-scale automated testing tools (Jest, pytest, GitHub Actions, Jenkins, commercial platforms like BrowserStack) routinely perform trial runs and validation in real organizations. AI-powered testing and static analysis tools are widely deployed and demonstrably reliable for catching common errors. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | Production tools (GitHub Copilot, AI-driven CI/CD test generation, autonomous coding agents) reliably run test suites and debug output today, though full autonomous validation of complex systems is still narrower in scope. |
Prepare detailed workflow charts and diagrams that describe input, output, and logical operation, and convert them into a series of instructions coded in a computer language.
84CI 81–86 · exposure 75 · augmentation 100 · importance 3.6/5 · click for rater detail
Prepare detailed workflow charts and diagrams that describe input, output, and logical operation, and convert them into a series of instructions coded in a computer language.
84| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 5/5 | Software development organizations have rapidly adopted AI coding tools; GitHub Copilot has millions of users, and LLM-based code generation is standard in many tech companies and enterprises. This is among the fastest-adopting sectors for AI automation. |
| Sector adoption velocity | claude-sonnet-5 | 5/5 | Software development is among the fastest and deepest AI-adopting fields, with coding assistants integrated into mainstream IDEs and enterprise workflows at scale. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI code generation and diagram-to-code tools dramatically augment programmer productivity by handling boilerplate, scaffold generation, and routine logic translation, allowing humans to focus on architectural decisions and testing while remaining actively in the loop. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI coding assistants dramatically speed up diagram creation, pseudocode drafting, and code translation while developers retain control over design decisions and final logic. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Current AI systems (e.g., GitHub Copilot, Claude, ChatGPT) can generate code from high-level descriptions and convert simple workflow diagrams into functional code, achieving significant time savings on routine implementations. However, complex architectural decisions and integration with legacy systems still require human oversight, preventing a full 5 rating. |
| Task automatability | claude-sonnet-5 | 4/5 | Modern LLM-based coding assistants can generate workflow diagrams (e.g., mermaid/flowcharts) and translate logic into working code from natural-language specs with substantial time savings, though complex legacy systems still need human validation. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Few legal or regulatory barriers exist to automating code generation itself. Organizations often impose internal review and testing requirements, but these are organizational friction rather than hard legal mandates; code produced by AI must still pass human testing and deployment controls. |
| Adoption barriers | claude-sonnet-5 | 1/5 | No licensing, legal, or human-contact requirement restricts who or what can draft workflow diagrams or write code; adoption is purely a technical/organizational choice. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | AI inference costs per code artifact are negligible (cents to sub-dollars) compared to the fully-loaded cost of a programmer's hourly wage ($75–150+ per hour), making the cost differential at least 10–100× in favor of AI. |
| Cost vs. human wage | claude-sonnet-5 | 5/5 | Inference costs for code/diagram generation are cents to dollars per task versus hourly programmer wages, giving at least an order-of-magnitude cost advantage even with oversight included. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Deployed products like GitHub Copilot, Codeium, and LLM-based coding assistants reliably generate code from natural language and structured descriptions in production environments. Error rates on simple-to-moderate complexity tasks are low, though hallucinations and incorrect logic occur on complex workflows. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | Deployed tools like GitHub Copilot, Cursor, and ChatGPT are widely used in production to draft diagrams and generate code from specifications, though outputs still require review for correctness and edge cases. |
Compile and write documentation of program development and subsequent revisions, inserting comments in the coded instructions so others can understand the program.
82CI 79–86 · exposure 75 · augmentation 100 · importance 3.6/5 · click for rater detail
Compile and write documentation of program development and subsequent revisions, inserting comments in the coded instructions so others can understand the program.
82| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Software development organizations, particularly in tech and finance, are rapidly adopting AI coding assistants for documentation and commenting in production workflows. Surveys show high pilot and early production adoption among information-sector firms. |
| Sector adoption velocity | claude-sonnet-5 | 5/5 | Software development is among the fastest and deepest AI-adopting fields, with AI coding assistants now standard in many professional environments. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI assistants dramatically augment programmer productivity when writing and maintaining documentation by generating first drafts, suggesting comments, and catching omissions—enabling developers to review and refine rather than author from scratch. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI dramatically speeds up documentation drafting and comment generation while programmers retain oversight to verify accuracy and context, a clear productivity transformation. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Current AI systems can automatically generate substantial code documentation and comments at scale with minimal human oversight, especially for well-structured code. While some nuance in domain-specific or legacy code explanation may require human review, the core task of inserting clarifying comments and writing technical documentation can easily exceed 50% time savings. |
| Task automatability | claude-sonnet-5 | 4/5 | LLM-based coding assistants can generate docstrings, comments, and changelogs from code with high accuracy, requiring only light human review, meeting the time-saving threshold for most routine documentation work. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Few legal or regulatory barriers prevent AI-assisted documentation; the main friction points are organizational preferences for human review and occasional liability concerns about inaccurate comments. No licensing requirement mandates human authorship of code comments. |
| Adoption barriers | claude-sonnet-5 | 1/5 | There is no licensing, liability, or regulatory requirement forcing a human to write code comments or documentation; adoption is purely a matter of workflow choice. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | AI inference costs for generating documentation and comments are negligible (fractions of a cent per task), making them orders of magnitude cheaper than professional developer time at typical loaded wages of $60-150/hour. |
| Cost vs. human wage | claude-sonnet-5 | 5/5 | Generating comments and documentation via AI costs fractions of a cent per function versus a programmer's hourly wage, making the cost differential very large. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Mature products like GitHub Copilot, ChatGPT, and Claude reliably generate code comments and documentation in production environments across many organizations. These systems demonstrate consistent performance on standard codebases, though error rates remain material for complex or unusual code patterns. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | Tools like GitHub Copilot, Cursor, and various IDE plugins reliably auto-generate comments and documentation in production use today, though complex architectural documentation still needs human refinement. |
Correct errors by making appropriate changes and rechecking the program to ensure that the desired results are produced.
80CI 77–82 · exposure 75 · augmentation 100 · importance 4.4/5 · click for rater detail
Correct errors by making appropriate changes and rechecking the program to ensure that the desired results are produced.
80| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 5/5 | Tech and fintech sectors are rapidly adopting AI-assisted debugging and error correction; Copilot and similar tools have millions of active users in production environments. Adoption is accelerating in high-digitization, information-intensive sectors. |
| Sector adoption velocity | claude-sonnet-5 | 5/5 | Software development is among the fastest and deepest AI-adopting fields, with AI coding assistants now standard in many professional workflows. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI substantially augments programmer productivity on this task by automating routine error detection, suggesting fixes, and running verification loops—allowing humans to focus on logic validation and architectural decisions. This is one of the most successful human-AI collaboration workflows in software development today. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially speeds up error diagnosis, suggests fixes, and helps verify results, transforming programmer productivity while the human retains final judgment and testing responsibility. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Current AI systems (GitHub Copilot, Claude, GPT-4) can identify and fix many common programming errors end-to-end with significant time savings. Automated testing, linting, and AI-driven debugging tools routinely catch and correct syntax, logic, and runtime errors, though complex architectural issues still require human judgment. |
| Task automatability | claude-sonnet-5 | 4/5 | Modern AI coding assistants can identify bugs, propose fixes, and iterate against test results with substantial time savings for many common error types, though complex logic errors and multi-file architectural bugs still require human oversight. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Few regulatory or licensing barriers exist for automating error correction in programming; code review and testing are already standardized practices. The main friction is organizational resistance to trusting automated fixes and the need for human verification in safety-critical systems. |
| Adoption barriers | claude-sonnet-5 | 1/5 | There is no licensing, regulatory, or liability barrier preventing use of AI for debugging code; organizations broadly permit and even encourage such tool use. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | AI-driven error correction (via subscription services or API calls) costs orders of magnitude less than human programmer time when amortized across codebases. Even with overhead for oversight and integration, the per-task cost is substantially lower than a programmer's loaded wage. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | AI-assisted debugging tools cost a small fraction (subscription fees of tens of dollars/month) compared to a programmer's hourly loaded wage, even accounting for the human review time still required. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Multiple deployed products (GitHub Copilot, Amazon CodeWhisperer, VS Code's IntelliSense, automated testing frameworks) reliably perform error detection and correction in production codebases. These tools are widely used in real organizations, though they still require human verification for complex or domain-specific errors. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | Production tools like GitHub Copilot, Cursor, and Claude Code are widely deployed and reliably assist with debugging and error correction in real engineering workflows, though they are not fully autonomous for complex systems. |
Write, analyze, review, and rewrite programs, using workflow chart and diagram, and applying knowledge of computer capabilities, subject matter, and symbolic logic.
76CI 69–82 · exposure 67 · augmentation 100 · importance 4.4/5 · click for rater detail
Write, analyze, review, and rewrite programs, using workflow chart and diagram, and applying knowledge of computer capabilities, subject matter, and symbolic logic.
76| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 5/5 | Tech and financial sectors show rapid, deep adoption of AI coding tools; surveys report 70%+ of developers using copilot-like systems, and enterprises are integrating AI agents into CI/CD pipelines at scale. |
| Sector adoption velocity | claude-sonnet-5 | 5/5 | Software development is among the fastest and deepest sectors for AI adoption, with coding assistants integrated into mainstream developer workflows at large scale already. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI dramatically augments programmer productivity by generating boilerplate, suggesting refactorings, spotting bugs, and accelerating code review—keeping humans in the loop while transforming output per unit time. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI coding assistants are widely reported to substantially boost programmer productivity in writing, reviewing, and rewriting code while the developer retains control and judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | AI can automate substantial parts of programming—code generation, refactoring, and static analysis—but end-to-end replacement falls short of the 50% time-saving bar at equal quality due to the need for human judgment on architecture, requirements validation, and complex debugging. |
| Task automatability | claude-sonnet-5 | 4/5 | Modern LLM-based coding assistants can generate, analyze, review, and rewrite substantial portions of code from specifications or diagrams with significant time savings, though complex system-level logic still needs human oversight.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Few legal or licensing barriers prevent AI-assisted programming; however, organizational friction (developer preference for control, security/IP concerns, code quality standards) and liability concerns over AI-generated bugs provide moderate friction to full automation. |
| Adoption barriers | claude-sonnet-5 | 1/5 | There is no licensing requirement or legal mandate for a human to write or review code; adoption is governed purely by quality/business considerations, not regulation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | AI inference and integration costs are very low (cents per task), while loaded programmer wages are $80–150k+ annually, making AI assistance substantially cheaper per unit of code reviewed or generated. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Subscription/API costs for AI coding tools (tens of dollars/month) are far below programmer wages for equivalent output on many subtasks, though integration and oversight costs reduce the ratio somewhat. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Mature products (GitHub Copilot, ChatGPT, Claude, specialized linters) reliably perform code generation, review, and some rewriting in production; however, workflow chart creation and full program analysis still require human oversight and sometimes struggle with novel or domain-specific logic. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | Deployed tools like GitHub Copilot, Claude, and Cursor are used daily in production by many programmers to write and refactor code reliably for well-scoped tasks, though full end-to-end program design still often needs human review. |
Write or contribute to instructions or manuals to guide end users.
75CI 66–84 · exposure 67 · augmentation 100 · importance 3.3/5 · click for rater detail
Write or contribute to instructions or manuals to guide end users.
75| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Software and tech companies have rapidly integrated AI-assisted documentation and code-generation tools into workflows; GitHub Copilot and similar systems are widely adopted in production environments. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software development is a fast-adopting, highly digitized sector where AI writing assistants are already widely integrated into IDEs and workflows. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI significantly augments programmer productivity for documentation tasks by generating drafts, examples, and boilerplate that humans then review, edit, and refine, allowing faster iteration while maintaining human oversight of quality. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially speeds up drafting, formatting, and updating manuals while programmers retain control over accuracy and final content. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | AI can draft technical documentation and generate code comments effectively, but typically requires significant human review, refinement, and validation of accuracy for production manuals. This covers roughly half the task with meaningful setup and oversight. |
| Task automatability | claude-sonnet-5 | 4/5 | LLMs can draft user documentation from code, comments, and specs with substantial time savings, though final review and accuracy checks are still needed for equal quality. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | There are few legal barriers to automation, though organizations often prefer human review for accuracy and liability concerns; professional judgment about tone and correctness creates modest friction but not hard legal requirements. |
| Adoption barriers | claude-sonnet-5 | 1/5 | No licensing, liability, or regulatory requirement mandates a human author manuals; organizations freely use AI-assisted drafting. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | AI-assisted documentation generation via inference is substantially cheaper than hiring technical writers or programmers to manually write comprehensive manuals, often saving 70–80% of labor costs even with quality review overhead. |
| Cost vs. human wage | claude-sonnet-5 | 5/5 | Generating draft documentation via AI costs a small fraction of a programmer's or technical writer's hourly rate for equivalent draft output. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Multiple AI products (GitHub Copilot, ChatGPT, specialized code documentation tools) demonstrably generate documentation and user guides today, though they often require human editing and fact-checking for accuracy and completeness. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | Deployed tools (Copilot, ChatGPT, dedicated doc-generation products) are routinely used in production to draft README files, API docs, and user guides today. |
Investigate whether networks, workstations, the central processing unit of the system, or peripheral equipment are responding to a program's instructions.
66CI 50–81 · exposure 58 · augmentation 88 · importance 3.3/5 · click for rater detail
Investigate whether networks, workstations, the central processing unit of the system, or peripheral equipment are responding to a program's instructions.
66| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 5/5 | High-digitization sectors (tech, finance, SaaS) have rapidly adopted automated monitoring, CI/CD pipelines, and AI-assisted diagnostics over the past five years. System health checks and hardware responsiveness monitoring are mainstream in DevOps practices. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | IT operations and DevOps sectors are moderately fast adopters of AI-assisted monitoring and diagnostics (AIOps tools), though full investigative automation for hardware-software interaction issues remains a pilot-stage practice in most organizations. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI dramatically augments programmer productivity by continuously monitoring systems in real time, flagging anomalies, and providing diagnostic traces—allowing humans to focus on root-cause analysis and remediation rather than manual log inspection and testing. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI significantly aids programmers by analyzing logs, suggesting likely failure points, and correlating symptoms across systems, substantially speeding up the investigative process even though a human directs and confirms findings. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Current AI systems can substantially automate the diagnosis of hardware responsiveness through log analysis, system monitoring, API calls, and automated testing frameworks. While some complex troubleshooting may require human judgment, the core investigative work—checking CPU usage, network connectivity, peripheral status, and comparing against expected behavior—can be performed end-to-end by AI agents with significant time savings. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can assist in diagnosing whether systems respond correctly by analyzing logs, error traces, and running diagnostic scripts, but complex multi-system debugging often requires human judgment about infrastructure context and business logic not fully captured by AI tools. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Few hard barriers exist; automation requires system access and integration with monitoring platforms but no licensing or legal restrictions prevent deployment. Some organizations may prefer human review for critical systems, but this is preference-based friction rather than regulatory requirement. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement blocks AI from analyzing system responsiveness, though enterprise IT environments often require human sign-off for infrastructure changes and root-cause conclusions due to liability and complexity. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | Automated monitoring and diagnostic tools operate at near-zero marginal cost per invocation compared to the loaded hourly wage of a programmer ($60–150+/hour). Once deployed, AI-driven system monitoring and responsiveness checking costs orders of magnitude less than manual investigation. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI diagnostic tools can reduce time spent on log analysis and pattern detection, offering moderate cost savings, but the need for human verification and system-specific knowledge keeps costs roughly comparable to skilled technician time in many cases. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Mature DevOps and system monitoring tools (e.g., Datadog, New Relic, automated testing frameworks) perform this task reliably in production environments today. AI-driven diagnostics via log analysis and system health checks are deployed at scale in organizations, though some edge cases and complex fault interactions may still require human verification. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Some AI-powered monitoring and diagnostic tools exist (e.g., anomaly detection, log analysis assistants) but full end-to-end investigation across networks, CPUs, and peripherals in production is still largely human-driven with AI as a support tool rather than an autonomous investigator. |
Perform or direct revision, repair, or expansion of existing programs to increase operating efficiency or adapt to new requirements.
63CI 57–69 · exposure 55 · augmentation 100 · importance 4.4/5 · click for rater detail
Perform or direct revision, repair, or expansion of existing programs to increase operating efficiency or adapt to new requirements.
63| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 5/5 | Software development is a high-digitization sector with rapid, measured adoption of AI coding tools; Copilot and similar tools see millions of developers in production use, with many enterprises piloting or deploying agent-based code generation and refactoring at scale. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software development is among the fastest-adopting sectors for AI tools, with copilots and agentic coding assistants integrated into everyday workflows at many tech companies. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI coding assistants dramatically augment programmer productivity by auto-completing code, suggesting refactorings, and accelerating bug detection and repair while keeping the programmer in full control. This is one of the clearest cases of transformative augmentation in the workplace today. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially speeds up code comprehension, bug identification, and refactoring suggestions, making it a strong productivity multiplier for programmers revising existing systems while they retain final control. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | AI can automate significant portions of code review, refactoring, and suggesting expansions via large language models, achieving meaningful time savings on routine maintenance tasks. However, directing complex architectural changes, understanding legacy system context deeply, and ensuring quality typically still require human oversight, preventing full end-to-end automation at the 50% threshold for complex programs. |
| Task automatability | claude-sonnet-5 | 3/5 | AI coding assistants can generate refactoring suggestions and identify inefficiencies, but the 'direct' aspect of revision requiring judgment about system architecture and requirements changes still needs substantial human oversight, especially for legacy or complex codebases.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Few legal or regulatory barriers prevent using AI for code revision, though organizational friction around code quality standards, security review, and version control practices provide some friction. Human programmer sign-off is generally expected but not legally mandated, leaving adoption relatively frictionless. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement blocks AI from performing code changes, though liability for bugs introduced into production systems and organizational code-review processes create moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | AI inference cost per code revision or repair is typically orders of magnitude cheaper than a programmer's loaded wage, though integration and human oversight still add cost that prevents a full 5 rating. For routine refactoring and bug fixes, the cost advantage is substantial. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI coding tools cost a fraction of a programmer's hourly wage per query, but the need for human review, debugging, and oversight of AI-suggested changes to production code narrows the effective cost advantage. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Mature products (GitHub Copilot, ChatGPT, Claude) demonstrably perform code revision and repair in production at scale across many organizations, with strong benchmarks on code generation and bug-fixing tasks. Some limitations remain on very large codebases or novel architectural decisions, but deployed systems reliably handle most routine program modifications. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Products like GitHub Copilot, Cursor, and Claude Code are deployed in production and materially speed up code revision and refactoring tasks, but reliability drops on large, complex, or poorly documented existing codebases requiring deep contextual understanding. |
Develop Web sites.
60CI 40–80 · exposure 50 · augmentation 88 · importance 3.2/5 · click for rater detail
Develop Web sites.
60| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Developer-heavy tech sectors and larger organizations are rapidly adopting AI coding assistants in production workflows; usage metrics show high penetration in software development, though as augmentation rather than replacement of the full task. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software development is among the fastest AI-adopting sectors, with widespread use of AI coding assistants and no-code/AI website builders already embedded in many workflows. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI coding assistants significantly boost developer productivity by generating boilerplate, suggesting implementations, and accelerating scaffolding; developers use these tools routinely to move faster while maintaining control over architecture and quality. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially boosts programmer productivity in web development through code generation, debugging, boilerplate creation, and design suggestions, while developers retain control over architecture and customization. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | AI can generate boilerplate code, simple layouts, and assist with styling, but end-to-end web development requires architectural decisions, integrations with backend systems, security hardening, and iterative refinement based on business logic that AI cannot reliably complete autonomously at production quality. |
| Task automatability | claude-sonnet-5 | 4/5 | Modern AI coding tools can generate functional websites (HTML/CSS/JS, frameworks, backend scaffolding) from natural language prompts, achieving significant time savings for standard site types, though complex custom architectures still require substantial human oversight. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Web development has low regulatory barriers and no licensing requirement; however, liability for security vulnerabilities, performance, and functional correctness creates organizational friction and requires human accountability, limiting pure automation. |
| Adoption barriers | claude-sonnet-5 | 1/5 | No licensing, regulatory, or legal requirement mandates a human programmer for building websites; adoption is purely a market and quality decision. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI coding assistance reduces per-line costs and speeds certain coding phases, but integration, testing, architecture review, and security validation still require skilled developers; total cost savings per complete website are modest relative to developer wages. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | AI-assisted web development tools cost a small fraction of a programmer's hourly wage and can produce basic-to-moderate complexity sites rapidly, though integration and QA oversight add some cost. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Code generation tools like GitHub Copilot and ChatGPT can produce snippets and templates, but no deployed product reliably builds entire functional websites from scratch without substantial human oversight, debugging, and architectural guidance. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | Products like GitHub Copilot, Cursor, Vercel v0, and various AI website builders (Wix ADI, Framer AI) are deployed in production and reliably generate working sites, though quality varies and human review is typically needed for complex requirements. |
Write, update, and maintain computer programs or software packages to handle specific jobs such as tracking inventory, storing or retrieving data, or controlling other equipment.
55CI 28–82 · exposure 50 · augmentation 88 · importance 4.2/5 · click for rater detail
Write, update, and maintain computer programs or software packages to handle specific jobs such as tracking inventory, storing or retrieving data, or controlling other equipment.
55| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Tech companies are adopting code generation tools in pilots and production settings, but adoption remains mixed and primarily assistive rather than replacement-level. Many organizations are still evaluating impact on productivity and quality rather than deploying AI for full maintenance automation. |
| Sector adoption velocity | claude-sonnet-5 | 5/5 | Software development is among the fastest and deepest AI-adopting fields, with AI coding assistants now standard in professional workflows across the tech industry. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI code assistants meaningfully augment programmer productivity by generating boilerplate, suggesting completions, and accelerating routine coding tasks, while programmers retain control over architecture, testing, and system design. This is one of the clearest cases of human-AI collaboration in knowledge work today. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI coding tools substantially speed up writing, updating, and debugging code while the programmer retains control over architecture, review, and integration decisions. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with code generation and refactoring, end-to-end writing, updating, and maintenance of production software systems requires domain knowledge, architectural decisions, testing, debugging, and ongoing lifecycle management that current AI systems cannot reliably perform without significant human oversight. The ≥50% time-saving bar is not consistently met across the full scope of this complex task. |
| Task automatability | claude-sonnet-5 | 4/5 | LLM-based coding tools can generate, update, and refactor substantial portions of standard business software (CRUD apps, inventory tracking, data storage/retrieval) with significant time savings, though complex system integration and debugging still require human oversight. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Substantial barriers exist: liability for bugs and security vulnerabilities in production systems, organizational risk tolerance, regulatory requirements in safety-critical domains, and the need for human architects to make design decisions. Organizations remain cautious about full automation of software maintenance. |
| Adoption barriers | claude-sonnet-5 | 1/5 | There is no licensing requirement or legal mandate for a human to write or maintain this kind of software; organizations are free to use AI-assisted or AI-generated code. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI coding assistants have low inference costs but require significant human review, testing, and correction overhead. The total cost of AI-assisted code generation plus human validation is not yet substantially cheaper than experienced programmer wages when quality and liability are factored in. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | AI coding assistants cost a small monthly subscription fee versus a programmer's hourly wage, and can produce boilerplate and standard modules far faster, though human oversight/integration time remains a nontrivial cost component. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Code generation tools (GitHub Copilot, ChatGPT) exist and are deployed, but they generate syntactically correct code that often requires substantial human review, testing, and debugging. Reliable end-to-end maintenance and updates of existing systems remain primarily human-driven; AI products handle fragments rather than the full task reliably. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | Products like GitHub Copilot, Cursor, and Claude Code are deployed at scale in production coding workflows and reliably generate functional code for common tasks, though they still require human review for correctness and edge cases. |
Consult with and assist computer operators or system analysts to define and resolve problems in running computer programs.
51CI 44–57 · exposure 42 · augmentation 88 · importance 3.5/5 · click for rater detail
Consult with and assist computer operators or system analysts to define and resolve problems in running computer programs.
51| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | AI coding assistants are already widely adopted in tech and financial services; many organizations routinely use Copilot and similar tools for debugging and problem-solving. Production adoption is measurable in information and professional services sectors, though human oversight remains standard. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software development and IT operations are among the fastest-adopting sectors for AI coding and debugging tools, with widespread pilot and production use of copilots. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI excels at augmenting this task: code analysis tools, error suggestion, and root-cause hypothesis generation meaningfully accelerate operator-programmer collaboration. The human remains central to understanding business context and approving fixes, but AI productivity gains are substantial. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI significantly augments programmers by quickly parsing logs, suggesting root causes, and drafting fixes, greatly speeding up the collaborative troubleshooting process while humans retain final judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist in debugging and problem diagnosis through code analysis, the task requires dynamic back-and-forth consultation with operators/analysts to understand context and nuance. AI cannot reliably drive the full diagnostic loop end-to-end while maintaining the collaborative problem-definition phase, making 50% time savings at equal quality unlikely without substantial human involvement. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can help diagnose bugs, interpret error logs, and suggest fixes, but the collaborative, real-time consultative nature with operators/analysts to define ambiguous problems still requires human judgment and context-gathering. |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Organizational friction exists: many firms require human sign-off on production system changes, operators may prefer human consultation for accountability, and liability concerns around automation of critical system fixes create friction. However, no explicit licensing mandate currently blocks AI-assisted diagnosis. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement, but organizational workflows, need for trusted judgment on production systems, and accountability for fixes create moderate friction against full substitution. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 3/5 | AI coding assistants have minimal inference cost, but the oversight overhead to verify correctness, integrate with specific system contexts, and validate fixes brings the all-in cost to roughly human-equivalent levels for high-stakes or complex problem resolution. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI tools reduce time spent on diagnosis and can be cheap per query, but human oversight, verification, and cross-team coordination still add meaningful cost, keeping totals roughly comparable for full task completion. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Current AI systems (copilots, code analysis tools) can suggest fixes and explanations for programming issues in deployed products, but they struggle with novel system configurations, implicit context from operators, and validation that a proposed solution actually resolves the stated problem without side effects. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Coding assistants and debugging copilots are deployed widely and help troubleshoot code, but they don't yet reliably replace the interactive consultation and problem-definition process with other staff. |
Train users on the use and function of computer programs.
44CI 38–50 · exposure 30 · augmentation 75 · importance 3.0/5 · click for rater detail
Train users on the use and function of computer programs.
44| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Many tech companies have adopted AI-assisted or automated training (e.g., interactive tutorials, knowledge bases, self-service learning), but live trainer displacement remains uncommon; hybrid models (AI + human) are more typical in professional settings. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Tech and software sectors are adopting AI-assisted documentation and chatbot support quickly, but formal end-user training remains a mixed adoption area with pilots more common than full deployment. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can assist trainers by drafting lesson plans, generating examples, creating documentation, and providing on-demand reference materials, substantially increasing instructor productivity while keeping the human expert in control of pacing and feedback. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can generate training materials, FAQs, scripts, and interactive tutorials, significantly boosting a trainer's productivity while they still lead sessions. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Training involves interactive explanation, assessing learner comprehension, and adjusting explanations—tasks requiring contextual understanding and real-time adaptation. While AI can generate training materials or provide scripted tutorials, current systems struggle with live, responsive instruction that handles diverse learning styles and unexpected questions. |
| Task automatability | claude-sonnet-5 | 2/5 | Training users involves live interaction, adapting to questions, reading confusion, and building rapport, which AI cannot fully replicate end-to-end despite generating training materials well.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Most organizations value human training for complex software, but there is no legal requirement mandating human instruction. However, user satisfaction, retention of tacit knowledge, and organizational preference for in-person onboarding create moderate friction against full automation. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement, but organizational preference for human trainers and the interactive, adaptive nature of training creates some friction against full substitution. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | AI-driven training platforms (video generation, chatbot support, self-paced modules) can be significantly cheaper than paying programmers to conduct live training sessions, especially at scale, though setup and oversight costs apply. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI can generate manuals or answer FAQs cheaply, live training still requires human facilitation, oversight, and customization, keeping costs comparable to or higher than pure AI content generation. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Deployed products like chatbots and automated training platforms can deliver basic tutorials and answer common questions reliably, but they often fail on complex or context-dependent queries and cannot replicate the rapport-building and motivation aspects of human instruction. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI chatbots and documentation generators exist to support training, but no deployed product independently conducts full user training sessions reliably in production. |
Perform systems analysis and programming tasks to maintain and control the use of computer systems software as a systems programmer.
42CI 28–57 · exposure 38 · augmentation 75 · importance 3.3/5 · click for rater detail
Perform systems analysis and programming tasks to maintain and control the use of computer systems software as a systems programmer.
42| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Tech and finance sectors are piloting AI-assisted code review and generation, but production displacement of systems programmers remains limited. Most adoption is augmentation (code suggestions, testing helpers) rather than autonomous task completion. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software and IT sectors are among the fastest adopters of AI coding tools, with widespread pilot and production use of AI-assisted development workflows. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI substantially assists systems programmers via code completion, automated testing, static analysis, and documentation generation, measurably raising productivity in implementation phases while humans retain control over architecture and deployment decisions. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI substantially boosts productivity for code generation, debugging, and documentation in systems programming work, while humans retain control over critical maintenance decisions. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Systems programming requires deep domain knowledge, architectural decision-making, and debugging of novel failures. While AI can assist with routine code generation and refactoring, end-to-end maintenance and control decisions involving system-level tradeoffs remain beyond 50% time savings at equal quality today. |
| Task automatability | claude-sonnet-5 | 3/5 | AI coding assistants can generate and modify code and assist with systems-level configuration, but maintaining and controlling production systems software requires ongoing judgment, integration with legacy environments, and accountability that current tools cannot fully replace end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Organizations face strong liability and operational risk barriers: automated systems changes can cause cascading failures affecting infrastructure, customers, and compliance. Regulatory frameworks and internal governance typically require human sign-off and accountability for critical systems changes. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No formal licensing is required for systems programmers, but organizational risk aversion around production system stability and security creates meaningful friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI coding assistance reduces cost per unit of boilerplate but systems programmers command high wages and require significant human oversight. Integration, testing, and validation of systems changes still demand experienced engineers, keeping all-in cost near parity with human labor. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI tools reduce time on coding subtasks significantly, but systems programming still requires substantial human oversight, testing, and validation, keeping all-in costs roughly comparable to a human-AI hybrid workflow rather than an order-of-magnitude cheaper. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | AI code generation tools (GitHub Copilot, Claude) exist but are applied to greenfield or well-documented tasks; systems programming maintenance requires understanding legacy code, system dependencies, and failure modes where AI reliability degrades significantly. Products do not reliably perform this task independently in production. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Deployed AI coding assistants (Copilot, Claude, etc.) are used in production for code generation and debugging, but autonomous systems programming/maintenance of critical systems software is not yet a mature standalone product category. |
Train subordinates in programming and program coding.
37CI 32–41 · exposure 25 · augmentation 75 · importance 3.5/5 · click for rater detail
Train subordinates in programming and program coding.
37| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Tech firms and consulting companies are piloting AI-assisted code review and learning platforms, but subordinate training remains primarily human-driven. Adoption is increasing in tooling (pair-programming assistants, automated code feedback) but has not yet displaced the human trainer role at scale. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Software/tech sector has relatively fast AI adoption overall, but the specific practice of using AI to train junior programmers is still emergent, with human-led mentorship models still dominant. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI significantly augments programmer-trainers through instant code review, automated documentation of common mistakes, and on-demand explanations of algorithms and APIs, allowing instructors to focus on high-level feedback, career guidance, and problem-solving strategy. Current tools (Copilot, ChatGPT, specialized code tutors) measurably raise instructor productivity in this domain. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI coding assistants and tutorials can significantly enhance a senior programmer's ability to train juniors by providing on-demand examples, explanations, and practice problems, substantially boosting training productivity while the human remains central to the process. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | AI can generate code examples and provide automated feedback on syntax or basic logic errors, but cannot replicate the adaptive, personalized guidance and mentoring that effective programmer training requires. The task involves understanding learner misconceptions, adjusting explanations in real-time, and building professional judgment—domains where AI lacks the contextual reasoning and relationship continuity needed for meaningful transfer of expertise. |
| Task automatability | claude-sonnet-5 | 2/5 | Training subordinates involves mentorship, assessing individual learning gaps, live feedback, and organizational context that AI cannot fully replicate end-to-end, though AI can supply supplementary materials and exercises.ed |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | While there are no strict legal barriers preventing AI-assisted instruction, organizational culture and manager accountability for team development create friction against full automation. Many firms view mentorship as a key leadership responsibility and retain preference for human judgment in evaluating readiness to advance. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No formal licensing requirement blocks AI-assisted training, but organizational preference for human mentorship, especially for junior staff development and team cohesion, creates moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 3/5 | An AI code tutor or automated feedback system costs relatively little to operate (per-user LLM inference + integration), but a programmer's training responsibilities typically represent a small fraction of their billable time, making direct replacement economically marginal; the value comparison is uncertain. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI-generated tutorials and code explanations are cheap, effective training requires ongoing human oversight, feedback loops, and relationship-building that still require senior programmer time, keeping costs comparable to human-led training. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | While AI tutoring products exist (code review assistants, interactive learning platforms), they operate best as self-directed supplements rather than autonomous trainers. Current systems struggle with diagnosing why a subordinate is struggling or pivoting explanations based on career-stage needs, making them poor substitutes for human instruction at scale in professional settings. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Products like AI coding tutors and documentation generators exist, but no deployed product reliably manages the full mentorship/training relationship between programmers and subordinates in production settings. |
Consult with managerial, engineering, and technical personnel to clarify program intent, identify problems, and suggest changes.
31CI 30–32 · exposure 25 · augmentation 63 · importance 4.0/5 · click for rater detail
Consult with managerial, engineering, and technical personnel to clarify program intent, identify problems, and suggest changes.
31| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Although software engineering is a high-digitization sector, actual adoption of AI agents for stakeholder consultation remains limited; most organizations still rely on human programmers for these coordination tasks. Early pilots exist, but production deployment for end-to-end consultation is rare. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Software development is a fast-adopting sector for AI coding tools, but the specific interpersonal consultation aspect of this task sees far more measured, assistive adoption rather than automation. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can usefully assist by generating meeting notes, surfacing known technical issues during discussions, or drafting change recommendations for human review, meaningfully raising a programmer's productivity in managing information during consultations while preserving human judgment on decisions. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can help programmers prepare talking points, summarize technical issues, translate requirements into plain language, and draft change proposals, meaningfully boosting productivity around the consultation. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can help draft summaries of conversations or suggest technical changes based on problem descriptions, it cannot reliably conduct the full consultative dialogue to clarify intent, identify problems through back-and-forth discussion, and propose context-aware changes in real-time with stakeholders. The task requires nuanced understanding of organizational dynamics and iterative human judgment. |
| Task automatability | claude-sonnet-5 | 2/5 | This task is fundamentally a live, multi-stakeholder communication and negotiation activity requiring real-time judgment about organizational priorities, not just code generation, so AI cannot fully replace it despite being useful for supporting materials. |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | While there are no strict legal barriers preventing AI support for internal consultations, organizational culture, the need for accountability in technical decisions, and stakeholder preference for direct human communication create moderate friction to full automation. Technical personnel often expect to speak with experienced engineers they can trust. |
| Adoption barriers | claude-sonnet-5 | 3/5 | There's no formal licensing requirement, but organizational trust, relationship dynamics, and accountability for judgment calls create real friction against replacing the human in these consultations. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | The AI overhead for real-time multi-stakeholder consultation systems (integration, fine-tuning, fallback oversight) remains high relative to the cost of human programmers conducting these conversations, which are often brief and episodic rather than continuous tasks. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Since a human programmer must still be present for the consultation itself, AI can only reduce prep/follow-up time modestly, so overall cost savings versus the human's loaded wage for this task are limited. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Current AI tools can assist in meeting analysis and suggest changes from transcripts or written descriptions, but no deployed product reliably conducts end-to-end consultative conversations with multiple stakeholders to clarify intent and identify problems independently. Products struggle with maintaining context across complex multi-party technical discussions. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI chat tools can help summarize requirements or draft clarifying questions, but no deployed product autonomously conducts stakeholder consultations to identify problems and negotiate changes in production settings. |
Assign, coordinate, and review work and activities of programming personnel.
29CI 25–32 · exposure 25 · augmentation 75 · importance 3.2/5 · click for rater detail
Assign, coordinate, and review work and activities of programming personnel.
29| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Although tech firms experiment with AI-assisted code review and task tracking, actual replacement of management and coordination roles remains rare; most deployments are assistive dashboards rather than agent-driven work assignment and oversight. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Software engineering teams are fast adopters of AI coding tools generally, but AI-driven management/coordination of personnel is still nascent and used mainly as an assistive layer rather than a replacement. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can meaningfully assist managers by automating routine task assignment, generating code review suggestions, surfacing team metrics, and flagging bottlenecks, thereby freeing the human manager to focus on mentoring, strategy, and complex coordination decisions. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist by summarizing code review comments, flagging code quality issues, tracking task status, and drafting performance feedback, boosting a manager's efficiency while they retain final judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Assigning work involves routine task distribution that could be partially automated, but reviewing work quality requires nuanced judgment about code architecture, team dynamics, and performance. End-to-end automation at 50% time savings is not reliably achievable with current AI. |
| Task automatability | claude-sonnet-5 | 2/5 | This is a managerial/coordination task involving people-management, prioritization, and review of subjective work quality; AI can support parts (ticket triage, code review comments) but cannot autonomously assign and manage human personnel end-to-end today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Team management and work review carry implicit accountability for code quality, security, and team performance; organizational structures, HR policy, and professional norms strongly favor a human manager bearing responsibility for personnel decisions and performance evaluation. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No formal licensing requirement, but organizational structure, accountability for team performance, and human trust in management roles create moderate friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Modest automation gains are possible, but the full managerial/coordinative task still requires significant human oversight and decision-making, keeping AI cost advantage limited compared to the salary cost of programming managers. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI tools reduce some overhead in tracking and reviewing code, but the managerial judgment, personnel coordination, and accountability still require a human at comparable or greater cost when factoring oversight. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with task assignment via scheduling tools and code review via static analysis, no mature product reliably handles the full coordinative and evaluative role that depends on real-time team context, performance history, and strategic priorities. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Products exist for project/task tracking (Jira, Linear) and AI code-review assistants, but no deployed system autonomously assigns and manages a team's work with reliable judgment about personnel and priorities. |
Collaborate with computer manufacturers and other users to develop new programming methods.
22CI 11–32 · exposure 13 · augmentation 63 · importance 2.8/5 · click for rater detail
Collaborate with computer manufacturers and other users to develop new programming methods.
22| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While AI coding tools see adoption in information sectors, strategic method development remains a human domain. Few organizations are replacing methodologists with AI; adoption is limited to narrow code-generation tasks, not method innovation itself. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Software development is a fast-adopting sector for AI coding tools generally, but this specific collaborative methods-development task sees little targeted AI adoption yet. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist by generating code examples, drafting documentation, or surfacing related techniques, helping humans iterate faster. However, the core work—conceptualizing, validating, and socializing new methods—remains a human leadership activity. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI coding assistants and research tools can help programmers explore new techniques, draft proposals, and analyze manufacturer specifications, meaningfully boosting productivity while humans lead the collaboration. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Developing new programming methods requires creative synthesis, domain expertise integration, and iterative refinement with stakeholders—tasks that demand human judgment and vision. Current AI cannot independently conceive and validate novel methodologies that advance the field. |
| Task automatability | claude-sonnet-5 | 2/5 | This task involves open-ended collaborative research and negotiation with external stakeholders to invent new methods, which requires creativity, relationship-building, and judgment that current AI cannot autonomously replace end-to-end.atable manner despite AI aiding sub-parts.RTHER |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Method development carries high organizational stakes—adoption, credibility, and liability concerns are significant. Stakeholder buy-in and expert validation are de facto requirements, creating strong friction against full automation or delegation to AI. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing requirement exists, but organizational trust, relationship history, and negotiation with manufacturers create real friction against AI substitution. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | The task involves senior-level expertise and sustained collaboration; human programmers with method-design experience command high wages. AI assistance is limited and cannot replace the core intellectual work, making total cost savings minimal. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Because AI cannot substitute for the collaborative and inventive core of this task, cost comparisons favor humans; AI can only cheaply support minor sub-tasks like documentation or code snippets. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product can autonomously develop new programming methods end-to-end. AI can assist in code generation and documentation, but strategic method development and cross-stakeholder collaboration remains fundamentally human-driven in practice. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product performs cross-organizational collaborative development of novel programming methodologies; this remains a human-driven R&D and partnership activity. |
Related occupations — Computer & Mathematical
How to read this
A high substitution score does not mean this job disappears — it means a large share of its current tasks face replacement pressure, so the mix of tasks is likely to change. High augmentation alongside substitution typically means the occupation reorganizes around the protected tasks. Wide confidence intervals mean the rater panel disagreed: treat those scores as open questions, not verdicts.
What would change this score
New model capabilities (automatability, feasibility), falling inference costs (cost ratio), regulation and licensing shifts (barriers), and measured sector adoption (velocity) all re-enter at every index release. Each release is recomputed, versioned and kept queryable — scores are claims with a date on them, not permanent labels.