Software Developers
15-1252.00Median wage $135,980/yr1,687,890 employed (US)Rank #108 of 923 scored · top 12% by substitution
Research, design, and develop computer and network software or specialized utility programs. Analyze user needs and develop software solutions, applying principles and techniques of computer science, engineering, and mathematical analysis. Update software or enhance existing software capabilities. May work with computer hardware engineers to integrate hardware and software systems, and develop specifications and performance requirements. May maintain databases within an application area, working individually or coordinating database development as part of a team.
Sub-scores
0–100 · band = confidence interval from rater disagreement
Substitution46
Exposure39
Augmentation78
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
12%
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%39
panel mean rating 2.6/5 → substitution pressure 39/100
Technical feasibility todayw 20%39
panel mean rating 2.6/5 → substitution pressure 39/100
Cost vs. human wagew 15%41
panel mean rating 2.7/5 → substitution pressure 41/100
Adoption barriersw 20%inverted — strong barriers lower the score61
panel mean rating 2.6/5 (barrier strength) → substitution pressure 61/100
Sector adoption velocityw 10%57
panel mean rating 3.3/5 → substitution pressure 57/100
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.
Prepare reports or correspondence concerning project specifications, activities, or status.
79CI 75–84 · exposure 75 · augmentation 100 · importance 3.7/5 · click for rater detail
Prepare reports or correspondence concerning project specifications, activities, or status.
79| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Software development and technology sectors show rapid adoption of AI writing tools; many organizations already use LLMs for documentation and email drafting as part of their standard workflows. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software/tech is among the fastest-adopting sectors for AI writing and documentation tools, with widespread integration into dev workflows. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI substantially augments developer productivity on this task by generating initial drafts, outlining specifications, and accelerating correspondence while developers retain editorial control and strategic input. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially speeds up drafting of reports and correspondence while developers retain control over content accuracy and final judgment calls. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | AI can generate initial drafts of status reports, specification summaries, and project correspondence with high accuracy using existing project data, meeting the ≥50% time-saving threshold. However, final review and context-specific refinement typically require developer judgment, preventing a perfect 5. |
| Task automatability | claude-sonnet-5 | 4/5 | Drafting status reports and correspondence from structured inputs (tickets, commit logs, specs) is well within current LLM capability, requiring mainly light human review/editing. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Few legal or regulatory barriers exist to AI-assisted report writing in software development. Light organizational friction may arise from preferences for human-authored correspondence or internal review policies, but nothing prevents substitution. |
| Adoption barriers | claude-sonnet-5 | 1/5 | No licensing, regulatory, or liability barriers prevent AI from drafting internal reports or correspondence. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | The marginal cost of AI inference for generating reports is negligible (cents per document) compared to a developer's loaded hourly wage (~$80–120/hour), creating an order-of-magnitude cost advantage. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Generating a report via an LLM costs a fraction of a developer's or PM's hourly wage, even accounting for review time. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Deployed products (ChatGPT, Claude, GitHub Copilot, and enterprise AI writing tools) reliably generate project documentation and emails in production settings. Some organizations use these at scale for routine correspondence, though quality varies with prompt specificity. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | Products like Copilot, Jira/Confluence AI add-ons, and general LLM assistants are already used in production to generate status updates and project documentation, though customization and accuracy checks are still needed. |
Store, retrieve, and manipulate data for analysis of system capabilities and requirements.
79CI 66–91 · exposure 67 · augmentation 100 · importance 3.6/5 · click for rater detail
Store, retrieve, and manipulate data for analysis of system capabilities and requirements.
79| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 5/5 | Software development is a high-digitization, information-sector domain with rapid adoption of AI-assisted tooling (GitHub Copilot, ChatGPT for code, automated data pipelines) in production environments. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software development is among the fastest-adopting sectors for AI coding tools, with widespread production use of AI-assisted coding and data-handling workflows. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI significantly augments developer productivity on this task through intelligent code completion, automated query generation, and schema recommendations, while developers remain in control of architectural decisions. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI coding assistants substantially speed up writing, debugging, and refactoring data storage/retrieval code, making this a strong augmentation use case while developers retain control over logic and validation. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Current AI systems can reliably handle most data storage, retrieval, and manipulation operations using SQL, APIs, and scripting with >50% time savings. However, novel schema design and complex analytical decisions require human oversight, preventing a full 5 rating. |
| Task automatability | claude-sonnet-5 | 3/5 | AI coding assistants can generate scripts and queries for data storage, retrieval, and manipulation, but integrating this into real system-capability analysis requires human framing, validation, and domain context that limits full end-to-end automation. |
| Adoption barriers | claude-haiku-4-5-20251001 | 1/5 | No licensing, regulatory, or legal requirement mandates human performance of data storage and retrieval tasks; adoption is purely economic and organizational, with no hard barriers to automation. |
| Adoption barriers | claude-sonnet-5 | 1/5 | There are no licensing or legal requirements restricting a developer or AI tool from performing data manipulation/analysis tasks; organizational risk tolerance is the main friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | AI-driven data manipulation via APIs, automated scripts, and code generation is orders of magnitude cheaper than hiring developers for routine data operations, with minimal human oversight needed. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | AI-assisted coding and query generation is inexpensive per unit of output compared to developer time, though human review and integration still add cost, keeping it just short of order-of-magnitude savings in most contexts. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 5/5 | Mature, production-grade systems (database management tools, ORMs, ETL platforms, and AI-assisted code generation) demonstrably perform these tasks at scale across organizations today with high reliability. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Deployed tools (Copilot, database query generators, data pipeline assistants) reliably help write and debug such code, but they still require developer oversight for correctness and system-specific integration, so reliability is narrow rather than fully autonomous. |
Monitor functioning of equipment to ensure system operates in conformance with specifications.
65CI 44–86 · exposure 55 · augmentation 75 · importance 4.1/5 · click for rater detail
Monitor functioning of equipment to ensure system operates in conformance with specifications.
65| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 5/5 | Observability and automated monitoring are now standard practice across high-digitization sectors (cloud, finance, tech, professional services). Adoption is deep and mature—the majority of modern software operations rely on automated system monitoring rather than manual checks. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software/tech sectors have rapidly adopted AI-powered observability, monitoring, and alerting tools, reflecting the fast adoption pattern typical of information/professional services industries. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI-powered monitoring significantly augments engineers' productivity by surfacing anomalies, correlating logs, suggesting root causes, and automating remediation decisions, allowing engineers to focus on investigation and resolution rather than manual log inspection and threshold checking. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI-based monitoring, anomaly detection, and alerting tools significantly enhance a developer's ability to track system conformance and specifications while the developer remains responsible for interpretation and action. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Current AI systems can reliably monitor equipment and log system states against specifications through automated alerting, log analysis, and anomaly detection—capturing 50–80% of traditional monitoring work. Human verification of complex failures remains necessary, but routine status checks and threshold violations are substantially automatable with off-the-shelf APM and observability tools. |
| Task automatability | claude-sonnet-5 | 2/5 | Monitoring equipment/system functioning against specifications requires continuous physical or real-time telemetry observation and judgment calls that current AI can assist but not fully replace end-to-end, especially when tied to physical equipment.jhs |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Monitoring automation faces only light barriers: no legal requirement for human sign-off, no licensing restrictions, and standard organizational acceptance of automated alerting. Some teams prefer hybrid oversight, but nothing prevents full substitution of routine monitoring. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement typically blocks automated monitoring, but liability for missed system failures and organizational reliance on engineer sign-off create moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | Cloud-based monitoring agents and services cost a small fraction of a software engineer's loaded wage (~$150–250/hour) for the amount of monitoring they handle continuously; cost per monitored metric is orders of magnitude lower than manual inspection labor. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | Automated monitoring tools (APM, observability platforms) are relatively cheap to run compared to constant human monitoring, but integration and tuning costs plus oversight needs keep the ratio moderate rather than an order of magnitude cheaper. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 5/5 | Deployed products (DataDog, New Relic, Prometheus, ELK Stack) demonstrably perform system monitoring at scale in production environments today, with integrated alerting and dashboarding. These are mature, widely adopted, and handle equipment and software system conformance monitoring reliably. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Monitoring dashboards and anomaly-detection tools exist and are deployed, but full autonomous conformance monitoring across diverse specifications is narrow and error-prone in production today. |
Obtain and evaluate information on factors such as reporting formats required, costs, or security needs to determine hardware configuration.
64CI 50–79 · exposure 58 · augmentation 88 · importance 3.1/5 · click for rater detail
Obtain and evaluate information on factors such as reporting formats required, costs, or security needs to determine hardware configuration.
64| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Software development and IT operations teams are rapidly adopting AI-driven code assistants and infrastructure-as-code tools that automate system configuration analysis. Major cloud platforms and DevOps workflows already embed AI recommendations for resource provisioning, indicating substantial and accelerating adoption. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Software development is a fast-adopting sector for AI tools generally, but this specific analytical/requirements-gathering task sees more pilot-stage use than full production deployment. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI significantly augments developer productivity by instantly surfacing relevant specifications, cost comparisons, and security checklists that would otherwise require manual research. Developers retain decision authority while AI transforms the speed and comprehensiveness of the evaluation phase. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can significantly speed up research, comparison of hardware options, and drafting of specification documents, greatly aiding the developer while they retain final judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Current AI systems can autonomously gather and analyze technical specifications, cost data, and security requirements from documentation and vendor resources, then synthesize recommendations with minimal human oversight. The task involves factual information retrieval and structured evaluation against known criteria, both well-suited to LLM and agent-based automation. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can gather and synthesize information on hardware requirements, costs, and security needs from documentation and specs, but determining an actual configuration requires integrating tacit organizational context and judgment that current systems only partially replicate.gs |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | There are minimal legal or regulatory barriers to automating this task; no licensing requirement mandates human sign-off on hardware recommendations. The main friction is organizational preference to retain human review for critical decisions and potential liability concerns, but these do not prevent automation. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No strict licensing requirement, but security and compliance considerations create moderate organizational risk aversion and need for accountable human sign-off. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | The cost of an API call or agent session to evaluate hardware requirements is orders of magnitude cheaper than paying a software developer's loaded wage for research and analysis work, especially when the task is routine and repetitive across projects. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI can cheaply produce draft analyses and comparisons, but human oversight and validation of security/cost tradeoffs is still needed, keeping costs roughly comparable once integration and review are factored in. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Deployed AI products (code assistants, documentation tools, and enterprise search systems) already handle specification analysis and configuration recommendation tasks in production settings. While some judgment calls about edge cases remain, mainstream AI tools reliably assist with and can largely automate this information-gathering and evaluation workflow. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Some AI-assisted tools can help analyze requirements and suggest configurations, but no mature production product autonomously performs this evaluative, multi-factor decision task reliably at scale. |
Modify existing software to correct errors, adapt it to new hardware, or upgrade interfaces and improve performance.
63CI 57–69 · exposure 55 · augmentation 100 · importance 3.7/5 · click for rater detail
Modify existing software to correct errors, adapt it to new hardware, or upgrade interfaces and improve performance.
63| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 5/5 | Software development is a high-digitization, information-sector occupation with rapid, measurable adoption of AI coding agents (Copilot installed on millions of machines, enterprise deployments growing). Public data and survey trends show strong, fast penetration. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software development is among the fastest-adopting professional domains, with AI coding assistants integrated into mainstream IDEs and workflows at major tech companies and beyond. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI code assistants demonstrably transform productivity for modification tasks—fast error detection, boilerplate generation, performance suggestions—while developers remain in control of architectural and business logic decisions. This is one of the canonical augmentation success cases. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially speeds up bug identification, code suggestion, and refactoring while developers retain control over testing, architecture decisions, and final validation. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | Software modification involves substantial human judgment in understanding code context, design intent, and trade-offs, but AI can automate significant portions—bug detection, boilerplate refactoring, performance profiling suggestions—achieving partial time savings. End-to-end automation remains limited due to need for architectural decisions and validation against complex requirements. |
| Task automatability | claude-sonnet-5 | 3/5 | AI coding assistants can locate bugs, suggest fixes, and generate patches for well-scoped issues, but complex legacy systems, architectural changes, and hardware adaptation still require substantial human judgment and testing. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Few regulatory or licensing barriers exist for AI-assisted code modification; organizations face mainly internal friction (code review, risk appetite, integration into CI/CD). No legal requirement mandates human authorship, and competitive pressure is low given industry standardization of AI tools. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement, but code review, testing, and deployment approval processes create organizational friction, and liability for introduced bugs or regressions encourages human oversight. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | AI coding tools cost pennies per task (inference) versus developer loaded wages ($50–150/hour); even with overhead and correction cycles, AI assistance is 10–50× cheaper per line of code produced. Integration and oversight costs remain modest. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI coding tools cost a fraction of developer salaries per query, but human review, testing, and integration overhead for correctness-critical fixes keeps the effective cost roughly comparable to accelerated human work rather than an order of magnitude cheaper. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Mature AI coding tools (GitHub Copilot, Claude, ChatGPT) demonstrably assist with code modification in production; they reliably handle error correction and interface updates on well-specified tasks. However, complex architectural changes or performance optimization still require human expertise, so feasibility is high but not universal. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Tools like GitHub Copilot, Cursor, and AI-assisted debugging are deployed in production and used daily by developers, but reliability drops sharply on large codebases, ambiguous bugs, or performance tuning requiring deep system knowledge. |
Develop or direct software system testing or validation procedures, programming, or documentation.
57CI 57–57 · exposure 50 · augmentation 100 · importance 3.9/5 · click for rater detail
Develop or direct software system testing or validation procedures, programming, or documentation.
57| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Software development is a high-digitization sector with fast AI adoption; major tech companies and development teams are already integrating AI code assistants and test generation tools into workflows at scale. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software development is among the fastest-adopting sectors for AI tooling, with widespread production use of AI coding assistants for testing and documentation tasks. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI assistants (Copilot, ChatGPT, specialized test-gen tools) significantly augment developer productivity for test writing, validation logic, and documentation—all core subtasks—while developers remain in control of strategic direction and validation procedures. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially accelerates writing test cases, boilerplate validation code, and documentation while developers retain control over test strategy and final validation logic. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | AI can automate parts of testing (test generation, execution of unit tests) and documentation (code documentation, API specs), achieving meaningful time savings. However, directing validation procedures and strategic testing decisions require human judgment, so end-to-end automation with 50%+ time savings is not reliable today. |
| Task automatability | claude-sonnet-5 | 3/5 | AI coding assistants can generate test cases, unit tests, and documentation drafts, but 'directing' testing procedures and validating complex system behavior still requires human architectural judgment and integration effort. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Regulatory and legal barriers are minimal; developers can adopt AI testing tools voluntarily. However, organizational friction exists around code quality standards and liability concerns if AI-generated tests miss critical cases, creating moderate adoption friction. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement for this task, though organizational code review and quality assurance processes create moderate friction before AI-generated artifacts are trusted in production. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 3/5 | AI inference costs for test generation and documentation are low, but integration and mandatory human oversight of generated tests/docs offset savings. The all-in cost approaches parity with developer time for these subtasks rather than achieving order-of-magnitude advantage. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI reduces time spent drafting tests/docs substantially, but the need for developer oversight, debugging AI-generated tests, and integration keeps overall cost roughly comparable rather than dramatically cheaper for the full task. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Mature products exist for code documentation generation (GitHub Copilot, ChatGPT) and test case generation, but they produce material errors and require significant human review. No deployed system reliably handles the full scope of directing testing procedures without supervision. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Products like GitHub Copilot, Claude, and specialized test-generation tools are deployed in production for generating tests and docs, but reliability varies and human review remains standard practice. |
Coordinate installation of software system.
47CI 40–54 · exposure 38 · augmentation 75 · importance 3.8/5 · click for rater detail
Coordinate installation of software system.
47| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | DevOps and software development sectors are among the fastest adopters of AI-assisted and automated tooling; CI/CD pipelines, orchestration, and deployment automation are widely deployed in production across high-digitization industries. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software/IT sectors have aggressively adopted DevOps automation, CI/CD, and AI-assisted deployment tooling, representing one of the faster-adopting professional domains. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI substantially augments installation coordination by automating environment provisioning, generating deployment manifests, monitoring rollouts, and suggesting fixes, meaningfully reducing cognitive load while developers retain final validation and troubleshooting control. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI-powered deployment assistants, chatops tools, and automated scripts significantly streamline coordination tasks like scheduling, logging, and troubleshooting, while a human remains in charge of final decisions. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | AI can automate significant portions of installation coordination—dependency resolution, configuration file generation, deployment scripting—but typically requires human oversight for environment-specific decisions, troubleshooting integration issues, and validation across heterogeneous systems, falling short of full end-to-end automation at equal quality. |
| Task automatability | claude-sonnet-5 | 2/5 | Coordination involves scheduling, cross-team communication, environment checks, and troubleshooting unexpected issues that require human judgment and organizational authority, limiting end-to-end automation today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Organizations often require human sign-off on production deployments for accountability and liability reasons, and many have policies or compliance requirements mandating human review, creating moderate friction against full automation. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement, but organizational risk tolerance for production installs, change-management policies, and accountability for failures create moderate friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | While AI reduces labor time on routine deployment tasks, the cost of infrastructure, monitoring, oversight, and human fallback for complex installations approaches or exceeds the cost of a developer performing the work directly, especially when failures are costly. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While automated deployment pipelines reduce some labor cost, the coordination overhead (communication, decision-making, exception handling) still requires human time comparable to or exceeding AI-assisted savings. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Products like Infrastructure-as-Code tools, deployment orchestration platforms (Kubernetes, Terraform), and emerging AI-assisted DevOps solutions exist and work reliably in controlled environments, but installation coordination in diverse legacy or custom environments still encounters material error rates and requires human judgment. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Deployment automation tools (CI/CD, orchestration platforms) exist and handle technical execution, but the 'coordination' aspect—aligning stakeholders, timing, rollback decisions—still relies on human project management in production settings. |
Design, develop and modify software systems, using scientific analysis and mathematical models to predict and measure outcomes and consequences of design.
42CI 28–57 · exposure 38 · augmentation 100 · importance 3.6/5 · click for rater detail
Design, develop and modify software systems, using scientific analysis and mathematical models to predict and measure outcomes and consequences of design.
42| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Software development is a digitized, early-adopting sector where AI coding tools see rapid uptake, but adoption is primarily for augmentation (code completion, debugging) rather than replacement of system design and architectural work. Deep, autonomous replacement of design work remains limited. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software development is among the fastest-adopting professional fields for AI tools, with widespread deployment of coding assistants and agentic tools in production environments across the tech sector. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI provides transformative productivity gains for software developers through code generation, testing suggestions, documentation, and performance analysis, while developers retain full control of architecture and design decisions. Augmentation is already widespread and proven in practice. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI coding assistants dramatically speed up code drafting, debugging, refactoring, and exploration of design alternatives, while developers retain control over final architecture and validation decisions. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with code generation and optimization, designing software systems using scientific analysis and mathematical modeling to predict outcomes requires substantial human judgment, architectural decision-making, and domain expertise that current AI cannot reliably perform end-to-end. AI tools support parts of implementation but cannot replace the full design cycle and validation process. |
| Task automatability | claude-sonnet-5 | 3/5 | AI coding assistants can generate substantial portions of code and even architectural suggestions, but end-to-end design involving novel scientific/mathematical modeling and validation of consequences still requires significant human judgment and iteration. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Organizational liability, regulatory compliance requirements (especially in safety-critical domains), and the necessity for human architects and technical leads to sign off on design decisions and validate mathematical/scientific soundness create substantial adoption friction. Professional responsibility and legal accountability strongly favor human ownership. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement blocks AI-assisted software design, though organizational risk tolerance, code review requirements, and liability for system failures create moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | While AI reduces certain implementation costs, the overhead of validation, redesign, and ensuring scientific/mathematical correctness means the all-in cost per task remains comparable to or potentially higher than a skilled developer's loaded wage, especially for mission-critical systems. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI tools reduce time on boilerplate and some design tasks but still require substantial paid developer oversight, integration, and validation, making the cost roughly comparable rather than order-of-magnitude cheaper for full task completion. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | AI code assistants (GitHub Copilot, ChatGPT) exist and help with implementation, but no deployed products reliably perform full system design, scientific analysis, and consequence prediction at production quality without extensive human oversight and revision. Current tools are narrow in scope and require skilled human judgment to validate outputs. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Products like Copilot, Cursor, and various AI coding agents are deployed widely and reliably assist with code generation and modification, but reliable autonomous handling of full system design with mathematical modeling of outcomes is not yet demonstrated at scale in production. |
Determine system performance standards.
42CI 32–51 · exposure 38 · augmentation 75 · importance 3.5/5 · click for rater detail
Determine system performance standards.
42| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Tech companies and cloud platforms have begun integrating automated performance monitoring and recommendation engines, but adoption remains concentrated in high-maturity shops. Broader small-to-medium enterprise adoption is still pilot-stage. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Software engineering as a field has moderate-to-fast AI tool adoption (e.g., Copilot-like assistants), though this specific sub-task—setting standards—sees only partial uptake via general LLM assistance. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI-powered performance profiling, anomaly detection, and recommendation engines meaningfully assist developers in identifying bottlenecks and simulating scenarios, substantially reducing manual instrumentation and analysis time while keeping the developer in control of decision-making. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can effectively help by researching benchmarks, industry norms, and drafting proposed thresholds, significantly speeding up the analysis phase even though final decisions stay human-driven. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | AI can assist with performance benchmarking, load testing configuration, and baseline metric identification, but determining standards requires understanding business requirements, cost-benefit trade-offs, and organizational constraints that typically need human judgment. Partial automation is feasible; full end-to-end replacement would fall short of the 50% time-saving bar. |
| Task automatability | claude-sonnet-5 | 2/5 | Defining performance standards requires business context, stakeholder negotiation, and judgment about tradeoffs that AI cannot independently originate; AI can draft suggestions but not own the decision end-to-end.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Organizational and technical friction exists: standards must align with business SLAs, regulatory compliance, and architectural choices that resist commoditization. However, no legal licensing barrier prevents automation of the technical measurement itself. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No formal licensing requirement, but organizational accountability and technical risk mean a responsible engineer or architect must approve standards, creating moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Full performance testing suites and integrations remain costly relative to developer labor, especially given the need for human validation and custom tuning per organization. AI inference itself is cheap, but total integration overhead makes the ratio unfavorable for routine tasks. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Because human judgment and stakeholder alignment remain necessary, AI mainly supplements rather than replaces the task, so cost savings are limited relative to the full task cost. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Tools exist (e.g., automated load testing frameworks, performance analysis platforms with AI-assisted recommendations) but they require significant human oversight to validate applicability and calibrate thresholds to actual system contexts. Deployable products perform narrowly, not holistically across heterogeneous environments. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | No deployed product autonomously sets performance standards for a system; existing tools only assist with benchmarking or suggesting metrics based on prior examples. |
Analyze information to determine, recommend, and plan installation of a new system or modification of an existing system.
41CI 34–49 · exposure 41 · augmentation 88 · importance 3.6/5 · click for rater detail
Analyze information to determine, recommend, and plan installation of a new system or modification of an existing system.
41| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While software development is a digital-first sector, AI adoption for system planning remains in the pilot and tool-assisted stage; most organizations still rely on human architects and senior developers for critical system design and installation decisions rather than AI-driven automation. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software development is a fast-adopting sector for AI tools, with widespread use of AI coding assistants and growing use in architecture and system design discussions in production settings. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can meaningfully augment human developers and architects by automating routine analysis, generating comparison matrices, surfacing compatible solutions, and drafting recommendation documents, while the human retains final judgment over system strategy and risk assessment. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially augments this task by helping developers research options, draft technical specifications, model trade-offs, and accelerate documentation, while humans retain decision-making authority. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | AI can assist with parts of system analysis and recommendation generation (e.g., comparing technical requirements, identifying compatible solutions), but planning the overall installation strategy—accounting for legacy dependencies, organizational constraints, risk assessment, and phased deployment—still requires human judgment and contextual knowledge beyond what current AI systems reliably deliver end-to-end. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can help gather requirements, compare options, and draft recommendations, but the core analysis of organizational constraints, stakeholder needs, and system trade-offs still requires significant human judgment and validation, limiting full end-to-end automation today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | System installation planning decisions carry significant liability and organizational risk; they typically require sign-off from licensed architects or senior engineers, and business stakeholders often require human accountability for technical recommendations affecting production systems. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No formal licensing requirement, but organizational risk aversion and the need for accountable technical ownership create moderate friction against fully automating this planning task. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | While AI can reduce time spent on documentation review and initial recommendation synthesis, the integration cost and need for human verification and high-stakes planning decisions mean overall cost remains comparable to or potentially higher than direct expert human analysis for complex, novel systems. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI can cheaply generate draft analyses, the need for senior engineer review, validation against business context, and integration oversight keeps overall costs closer to human-comparable rather than order-of-magnitude cheaper. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Some tools exist to support requirements analysis and generate implementation recommendations (e.g., code analysis, architecture pattern matching), but no deployed product reliably performs the full analysis-to-planning pipeline for novel system modifications at production quality without significant human oversight and domain expertise. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Deployed AI coding/architecture assistants (e.g., Copilot, ChatGPT-based tools) can suggest system designs and analyze code, but no production system reliably performs full system analysis and installation planning without extensive human oversight. |
Train users to use new or modified equipment.
37CI 32–41 · exposure 25 · augmentation 75 · importance 2.9/5 · click for rater detail
Train users to use new or modified equipment.
37| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Tech and larger enterprises are experimenting with AI-generated training content and chatbot support, but production deployment remains limited; most organizations still rely on human-led training for equipment onboarding. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Software/tech sectors are fast adopters of AI tools generally, but the specific sub-task of live user training is still commonly human-led, with AI adoption limited mostly to supporting materials. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can substantially assist trainers by generating curriculum drafts, creating interactive simulations, answering repetitive FAQs, and personalizing learning paths, allowing humans to focus on hands-on demonstration and adaptive mentoring. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can significantly boost training efficiency by drafting materials, generating FAQs, creating interactive tutorials, and answering routine user questions, while a human remains involved for complex training delivery. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Training requires adaptive communication, real-time response to learner confusion, and contextual pedagogical judgment. While AI can generate training materials and scripts, end-to-end training with equipment interaction demands human presence and responsiveness that current systems cannot reliably replicate at quality parity. |
| Task automatability | claude-sonnet-5 | 2/5 | Training users involves live demonstration, adaptive explanation, and hands-on troubleshooting tailored to individual users, which current AI cannot fully replace end-to-end despite being able to generate training materials or scripts.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Organizations typically prefer or require human trainers for accountability, liability, and user confidence; however, no explicit licensing barrier exists, and many companies are piloting AI-assisted training materials. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement, but organizational preference for human trainers for complex or safety-critical equipment, plus need for real-time adaptive interaction, creates moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI-assisted training materials have low per-unit cost, but oversight, refinement, and human trainer involvement remain necessary; the total cost of AI-supplemented training is often comparable to or higher than direct human training for small cohorts. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI can cheaply generate training content and answer FAQs, but live instructor-led sessions and hands-on support still require human time, keeping overall costs roughly comparable. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Chatbots and LLM-based tutoring systems exist but have narrow scope (text-only, limited interaction), high error rates in debugging user problems, and cannot physically demonstrate or intervene. No mature production system reliably trains users on modified equipment across diverse scenarios. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Some products (chatbots, AI-generated tutorials, documentation assistants) support parts of user training, but no deployed system reliably conducts full equipment training sessions in production at scale. |
Confer with systems analysts, engineers, programmers and others to design systems and to obtain information on project limitations and capabilities, performance requirements and interfaces.
31CI 25–38 · exposure 25 · augmentation 63 · importance 3.8/5 · click for rater detail
Confer with systems analysts, engineers, programmers and others to design systems and to obtain information on project limitations and capabilities, performance requirements and interfaces.
31| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While tech companies use AI for documentation and ideation, actual adoption of AI-led design conferences remains experimental and limited. Most software teams still heavily rely on human-driven collaboration; AI is supplementary rather than replacing the conferencing task itself. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Software development is a fast-digitizing profession increasingly using AI copilots and meeting assistants, though the specific interpersonal requirements-gathering activity sees more pilot use than full production automation. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can meaningfully assist by preparing requirement summaries, generating meeting agendas, drafting architectural sketches, and identifying gaps in specifications before or after meetings. These augmentations improve efficiency and capture but do not transform the core collaborative task, which remains human-driven. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI tools can significantly aid this task by transcribing discussions, summarizing requirements, generating diagrams, and drafting documentation, meaningfully boosting the developer's productivity while humans remain central to the interaction. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can draft meeting notes, summarize requirements, and generate documentation, the collaborative design process requires real-time problem-solving, creative synthesis of diverse constraints, and interpersonal negotiation that current AI cannot handle end-to-end. An AI could assist with preparation and documentation but cannot replace the core conferencing and decision-making. |
| Task automatability | claude-sonnet-5 | 2/5 | This task is fundamentally interpersonal—synthesizing stakeholder input, negotiating constraints, and building shared understanding across teams—which current AI cannot autonomously conduct end-to-end., though AI can summarize or draft notes from such meetings. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Technical design conferences involve judgment calls, accountability for system architecture decisions, and cross-functional stakeholder alignment that organizations and individuals are reluctant to delegate fully to AI. Liability for poor design decisions, regulatory sign-off requirements, and the expectation that engineers own their systems create strong adoption friction. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing barrier exists, but organizational and human-contact norms mean stakeholders expect to confer with human designers/engineers, creating moderate friction against full substitution. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Current AI can reduce overhead (transcription, note prep) but cannot eliminate the need for human facilitators and decision-makers. The cost of integrating AI tools for meeting support, combined with ongoing human involvement, approaches or exceeds the savings from partial automation. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Human collaboration and judgment remain necessary for this task, so AI mainly adds tooling cost on top of the human's time rather than substituting for it, keeping cost savings modest. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product reliably conducts multi-stakeholder technical design conferences independently. AI tools can participate in meetings (transcription, note-taking) and suggest solutions to isolated problems, but orchestrating and mediating complex technical discussions among humans remains a narrow, unreliable capability in practice. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Products like meeting transcription/summarization tools and AI chat assistants exist to support parts of this exchange, but no deployed system independently confers with stakeholders to extract requirements and constraints reliably. |
Consult with customers or other departments on project status, proposals, or technical issues, such as software system design or maintenance.
29CI 25–32 · exposure 25 · augmentation 75 · importance 3.4/5 · click for rater detail
Consult with customers or other departments on project status, proposals, or technical issues, such as software system design or maintenance.
29| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While software development teams use collaboration tools, actual replacement of developer-customer consultation with AI remains rare in production; adoption is primarily in support tasks rather than primary consulting. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Software industry has fast AI adoption overall, but this specific consultative, relationship-based task lags behind more automatable coding tasks within the same occupation. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI excels at preparing briefing materials, generating status summaries, suggesting technical solutions, and documenting decisions—significantly boosting a developer's ability to consult effectively while the human maintains relationship and accountability. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI tools can meaningfully assist by drafting status reports, summarizing technical issues, preparing talking points, and generating documentation to support the human-led consultation. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can draft status updates or summarize technical information, this task requires real-time dialogue, understanding evolving customer needs, and building trust—elements that demand human judgment and accountability. AI might assist in preparation but cannot reliably conduct the full consultation end-to-end. |
| Task automatability | claude-sonnet-5 | 2/5 | This task centers on real-time interpersonal consultation, negotiation, and reading stakeholder needs, which AI cannot yet fully replicate end-to-end despite handling some information retrieval or summarization sub-components. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Customer and stakeholder relationships carry legal, reputational, and business risk if automated; most organizations require a human developer to be accountable for technical commitments and status claims. Client expectations and contractual obligations typically mandate human involvement. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing requirement, but strong organizational friction exists since customers expect a human point of contact for accountability, trust, and nuanced negotiation on technical/project matters. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI systems require setup, fine-tuning for domain specifics, and human oversight to prevent miscommunication or technical misstatement. The loaded cost of a developer's time and the cost of AI-driven errors often exceeds the savings from partial automation. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Human consultation requires relationship management and accountability that AI cannot cheaply replace; AI can lower prep costs but the core interaction still requires a paid human presence. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Chatbots and AI agents can provide scripted responses and generate technical summaries, but deployed systems lack the contextual reasoning, relationship management, and error recovery needed for genuine cross-departmental consulting in production environments. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | No deployed product autonomously conducts client/department consultations on technical issues; AI chatbots exist for narrow support but not for substantive project-status or design consultations with judgment and trust involved. |
Confer with data processing or project managers to obtain information on limitations or capabilities for data processing projects.
29CI 25–32 · exposure 25 · augmentation 63 · importance 3.3/5 · click for rater detail
Confer with data processing or project managers to obtain information on limitations or capabilities for data processing projects.
29| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Despite high digitization in tech, actual delegation of stakeholder conferencing to AI agents remains rare; most organizations still insist on developer-manager direct communication for these exploratory, judgment-laden conversations. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Software development is a high-digitization field with fast AI tool adoption generally, but this specific interpersonal coordination task lags behind code-generation and technical adoption within the same occupation. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist by preparing summaries of past project constraints, suggesting clarification questions, drafting meeting notes, and surfacing relevant documentation before or after the conference, meaningfully accelerating the process while the developer remains the primary agent. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist by preparing briefing materials, summarizing prior documentation, drafting questions, and capturing/synthesizing notes from the conversation, improving efficiency while the human still conducts the actual conferring. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can retrieve and summarize project documentation or create meeting agendas, the task fundamentally requires real-time dialogue, stakeholder understanding, and contextual judgment to extract nuanced constraints and capabilities. Current AI cannot reliably replace the interactive clarification and relationship-building essential to effective conferencing. |
| Task automatability | claude-sonnet-5 | 2/5 | This is a synchronous interpersonal conversation requiring negotiation, context-gathering, and reading organizational nuance, which current AI cannot conduct autonomously end-to-end.5AI chat tools can support parts (e.g., summarizing prior docs) but cannot replace the actual conferring interaction. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Organizational norms, stakeholder preference for direct human communication, and the implicit requirement that the developer understand and own the constraints create strong friction against full automation. Managers expect to speak with the actual developer, not an AI proxy. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing barrier, but organizational and interpersonal trust dynamics create real friction—managers expect to interact with a human developer who understands institutional context and stakes. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Deploying AI to autonomously conference with managers would require extensive integration, override of confidentiality concerns, and human verification of outputs, making it comparable to or more expensive than a developer simply attending the meeting themselves. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Since AI cannot substitute for the actual human conferring, cost comparison is largely moot; any AI use is a supplement layered onto existing meeting/labor costs rather than a replacement, so no significant cost savings are realized on the core task. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | AI chatbots and meeting assistants exist but cannot independently conduct these conferences; they lack the authority to commit information, the social credibility to extract honest constraints, and the ability to navigate complex organizational politics without human presence. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | No deployed product autonomously conducts these managerial requirements-gathering conversations; existing tools are limited to note-taking, transcription, or summarization support rather than conducting the exchange itself. |
Analyze user needs and software requirements to determine feasibility of design within time and cost constraints.
28CI 14–42 · exposure 20 · augmentation 38 · importance 4.1/5 · click for rater detail
Analyze user needs and software requirements to determine feasibility of design within time and cost constraints.
28| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Tech companies are piloting AI for requirements analysis and early design support, but production deployment remains limited; many organizations still rely on human expertise for feasibility gates, though augmentation tools are becoming more common. |
| Sector adoption velocity | claude-sonnet-5 | 1/5 | placeholder |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI tools can significantly boost developer productivity by rapidly extracting requirements from documents, highlighting implicit constraints, and generating feasibility checklists; developers using assistive AI for this task can move faster while retaining final judgment authority. |
| Augmentation potential | claude-sonnet-5 | 1/5 | placeholder |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | AI can assist with some parts of requirements analysis—document parsing, identifying explicit constraints, listing features—but feasibility determination requires judgment about system design tradeoffs, stakeholder priorities, risk assessment, and integration with existing systems that demand human expertise and domain knowledge. |
| Task automatability | claude-sonnet-5 | 2/5 | This task requires synthesizing ambiguous stakeholder input, organizational context, and business constraints into a feasibility judgment, which current AI cannot do end-to-end reliably; it can support parts (drafting requirement docs) but not the core judgment.6 |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | This task typically requires developer accountability for feasibility assessments in production environments; liability for missed constraints or unfeasible commitments, plus the need for human sign-off and architect review, creates substantial friction against full automation. |
| Adoption barriers | claude-sonnet-5 | 1/5 | placeholder |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI services for requirements analysis are still relatively expensive per task when accounting for integration and mandatory human oversight; the cost approaches but does not yet undercut a mid-level developer's hour for this complex analysis work. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | placeholder |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | While AI can summarize requirements documents and flag constraints, no deployed product reliably performs end-to-end feasibility analysis independently; current tools require substantial human review of AI output to validate architectural decisions and realistic cost/time estimates. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | placeholder |
Supervise the work of programmers, technologists and technicians and other engineering and scientific personnel.
16CI 7–25 · exposure 13 · augmentation 50 · importance 3.5/5 · click for rater detail
Supervise the work of programmers, technologists and technicians and other engineering and scientific personnel.
16| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Adoption of AI for supervisory roles is minimal in practice; tech companies experiment with AI-assisted dashboards for managers, but actual automation of supervision decisions is laggard due to risk aversion and the strategic importance of human leadership. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Even in fast-adopting tech sectors, AI adoption is concentrated in coding assistance, not in replacing management/supervisory functions over staff. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can meaningfully assist supervisors by aggregating team metrics, flagging performance anomalies, scheduling, and surfacing data for decision-making, raising management productivity on administrative and analytical aspects while the human retains judgment and accountability. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help supervisors track project status, review code quality, summarize team progress, and flag issues, offering moderate assistance while the human remains the decision-maker and people manager. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Supervising technical staff requires real-time human judgment, relationship management, and context-dependent decision-making about complex work. While AI could handle routine status updates or escalation flagging, the core supervisory task—coaching, conflict resolution, performance evaluation, and strategic project decisions—remains largely beyond current AI capabilities and requires human accountability. |
| Task automatability | claude-sonnet-5 | 1/5 | Supervising people involves personnel management, mentoring, and accountability that current AI systems cannot perform end-to-end; no time-saving automation of the core supervisory act exists today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Strong organizational and legal barriers exist: supervisory authority typically requires human judgment, liability for team decisions rests on a licensed or responsible human, and most organizations would face substantial friction replacing human managers with autonomous systems due to employment law and accountability requirements. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Supervisory responsibility carries organizational accountability, legal/HR liability, and requires human judgment and authority that firms are unlikely to delegate to software, though it's not formally licensed like medicine or law. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | The overhead of human oversight, correction, and liability for AI-driven supervision decisions would likely exceed the cost of straightforward human supervision, especially given the seniority and accountability burden of the role. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no AI substitute performing this task, so cost comparison favors the human by default; AI cannot replace the managerial function being priced. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No production systems reliably perform end-to-end technical supervision today. AI can assist with meeting notes, performance metrics aggregation, or schedule coordination, but deployed products fail at the nuanced interpersonal and accountability aspects that define supervision. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product manages or supervises human engineering staff; AI tools assist with code review or task tracking but do not perform supervisory oversight of personnel. |
Supervise and assign work to programmers, designers, technologists, technicians, or other engineering or scientific personnel.
16CI 7–25 · exposure 13 · augmentation 63 · importance 3.1/5 · click for rater detail
Supervise and assign work to programmers, designers, technologists, technicians, or other engineering or scientific personnel.
16| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Supervisory and team-management tasks remain heavily human-centered even in tech organizations; while project management tools proliferate, autonomous AI assignment of work to personnel is rare in production, and adoption is limited to pilots or assist-only scenarios. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | While software/tech sectors adopt AI tools quickly for coding tasks, actual delegation of people-management and work assignment to AI remains rare and experimental. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist supervisors by suggesting workload balancing, tracking capacity, or flagging scheduling conflicts, improving visibility and responsiveness; however, the assistance is bounded by the need for human judgment on priorities and personnel fit. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can help managers track project status, suggest task assignments based on skills/workload, and draft schedules, meaningfully assisting the human decision-maker. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Assigning work requires judgment about individual capabilities, project priorities, and interdependencies that current AI systems struggle with reliably. While AI can help draft assignments or suggest task allocation, the supervisory responsibility and need for contextual understanding of team dynamics limits end-to-end automation to below the 50% time-saving threshold. |
| Task automatability | claude-sonnet-5 | 1/5 | Supervising and assigning work to human staff requires interpersonal judgment, accountability, mentorship, and organizational authority that current AI systems cannot exercise end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Supervision of personnel and work assignment carries organizational, legal, and accountability requirements; human managers are typically expected to own personnel decisions, and regulatory/HR frameworks generally require human judgment and sign-off on work allocation and performance oversight. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Supervisory responsibility typically requires a human manager accountable for personnel decisions, performance evaluation, and organizational hierarchy, creating strong structural and liability barriers. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Current AI tools (copilots, scheduling assistants) are relatively inexpensive but still require human supervision and decision-making, so the total cost-per-assignment remains comparable to or higher than a manager's time investment in this task. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no viable AI substitute performing this managerial function, so cost comparison favors the human who retains legal and organizational responsibility. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No production systems demonstrably perform full supervisory assignment end-to-end; AI tools can assist with scheduling or workload tracking but lack the judgment and accountability for meaningful personnel oversight that deployed products handle reliably. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously manages or assigns human personnel work; project management tools assist but don't perform supervisory judgment or accountability. |
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.