Computer and Information Research Scientists
15-1221.00Median wage $140,300/yr37,200 employed (US)Rank #479 of 923 scored · top 52% by substitution
Conduct research into fundamental computer and information science as theorists, designers, or inventors. Develop solutions to problems in the field of computer hardware and software.
Sub-scores
0–100 · band = confidence interval from rater disagreement
Substitution28
Exposure23
Augmentation71
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
15 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
0%
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%24
panel mean rating 2.0/5 → substitution pressure 24/100
Technical feasibility todayw 20%22
panel mean rating 1.9/5 → substitution pressure 22/100
Cost vs. human wagew 15%22
panel mean rating 1.9/5 → substitution pressure 22/100
Adoption barriersw 20%inverted — strong barriers lower the score39
panel mean rating 3.4/5 (barrier strength) → substitution pressure 39/100
Sector adoption velocityw 10%38
panel mean rating 2.5/5 → substitution pressure 38/100
Task breakdown (15 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.
Participate in staffing decisions and direct training of subordinates.
42CI 3–82 · exposure 45 · augmentation 63 · importance 3.0/5 · click for rater detail
Participate in staffing decisions and direct training of subordinates.
42| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Large tech, finance, and professional services firms rapidly adopt AI-driven recruitment and training platforms; mid-market adoption is accelerating. The information sector where this task is most prevalent shows fast, measurable displacement of manual screening and training workflows. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | While HR tech and analytics tools are gradually adopted in tech/research organizations, the core people-management decisions remain human-led with only slow, cautious integration of AI-assisted tools. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI assists managers by automatically ranking candidates, flagging skill gaps, and generating personalized training recommendations, freeing the manager to focus on final selection judgment and mentorship—substantially raising manager productivity while keeping humans in strategic decisions. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can assist with resume screening, training material generation, and performance data analysis, providing moderate support to a manager who retains ultimate responsibility for staffing and training decisions. |
| Task automatability | claude-haiku-4-5-20251001 | 5/5 | Current AI systems can fully automate recruitment candidate screening, resume ranking, skills-gap analysis, and personalized training plan generation with substantial time savings and consistent quality. The task requires no inherently human judgment that AI cannot replicate at or above human performance levels. |
| Task automatability | claude-sonnet-5 | 1/5 | Staffing decisions involve interpersonal judgment, organizational politics, and legal/HR considerations that current AI cannot execute end-to-end; AI cannot legitimately hire, fire, or direct human subordinates autonomously.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | While no hard licensing barrier exists, organizational policies and risk-aversion around delegating final hiring decisions to AI create moderate friction. Employment law compliance, reputational concerns about algorithmic bias, and human preference for manager-driven staffing create meaningful but surmountable adoption resistance. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Hiring and personnel management carry strong legal, employment-law, and liability requirements (discrimination law, labor regulations) that require human accountability and sign-off, making this a highly protected managerial function. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | AI-driven recruitment screening, skill assessment, and adaptive training platforms cost orders of magnitude less per hire/trainee than the fully-loaded salary of a manager performing these duties. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no viable AI substitute performing this managerial task alone, so cost comparison is moot; any AI use is supplementary decision-support, not a replacement of the human labor cost. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Mature HR platforms (LinkedIn Recruiter, Workable, Cornerstone OnDemand) and learning management systems already perform candidate ranking and training assignment at scale. Some edge cases around hiring decisions remain reliant on human review, but core task performance is demonstrably reliable in production. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product independently makes staffing decisions or directs employee training; HR software offers analytics/screening support but a human manager retains and exercises the decision authority. |
Assign or schedule tasks to meet work priorities and goals.
38CI 30–46 · exposure 38 · augmentation 75 · importance 3.8/5 · click for rater detail
Assign or schedule tasks to meet work priorities and goals.
38| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Research institutions and tech companies show slow-to-middling adoption of fully automated task scheduling; while project management tools are widespread, autonomous priority-driven assignment remains in pilot phase, with most organizations retaining human scientist oversight on critical task allocation. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Tech and R&D organizations are moderately fast adopters of AI-assisted project management tools, though full automation of task assignment remains uncommon and mostly pilot-stage. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can substantially augment this task by auto-generating candidate schedules, surface resource conflicts, flag priority inconsistencies, and handle routine assignments, leaving humans to make final decisions—significantly raising the speed and quality of task planning without removing human judgment from strategic allocations. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist by analyzing workloads, suggesting priorities, and optimizing schedules, significantly aiding a manager who retains final decision-making authority. |
| Task automatability | claude-haiku-4-5-20251001 | 3/5 | Task assignment and scheduling can be partially automated through AI systems that analyze project requirements, resource availability, and priorities, but requires human judgment on nuanced priority trade-offs, team dynamics, and goal alignment—typically delivering 30-50% time savings with significant setup needed. |
| Task automatability | claude-sonnet-5 | 2/5 | Task assignment involves judgment about team members' skills, workload, motivation, and organizational priorities that AI cannot yet fully assess or execute end-to-end without significant human oversight.deleted |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Task assignment in research settings has moderate adoption friction: organizations value human oversight of prioritization, team expertise matching requires contextual judgment, and research culture often favors human decision-making on goal-critical allocations, creating organizational resistance rather than legal barriers. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing requirement, but organizational trust, managerial authority, and accountability for team performance create moderate friction against full delegation to AI. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | The loaded cost of a research scientist's time for this task is substantial; while AI scheduling tools have low per-task inference costs, integration, configuration, and human oversight required to ensure quality assignments still makes the all-in cost comparable to or higher than direct human assignment. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While software tools are cheap, the managerial judgment required for effective task assignment still requires substantial human time, so AI alone doesn't yield large cost savings for this specific task. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Products like AI-assisted project management tools and workflow schedulers exist in production, but they often require manual intervention, oversight, and correction; full autonomous task assignment without human review remains unreliable at scale in complex research environments. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Project management tools offer scheduling assistance and some AI-driven prioritization features, but no deployed product autonomously assigns tasks to research staff reliably in production without manager review. |
Analyze problems to develop solutions involving computer hardware and software.
34CI 28–40 · exposure 25 · augmentation 88 · importance 4.2/5 · click for rater detail
Analyze problems to develop solutions involving computer hardware and software.
34| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Tech companies are actively experimenting with AI-assisted development and analysis (GitHub Copilot adoption, internal AI tooling), but deployment remains primarily assistive rather than autonomous. Adoption is growing in research and development pipelines but has not reached deep displacement of the core analysis function. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software/tech research and engineering sectors are among the fastest adopters of AI coding and analysis tools, with widespread pilot and production use of AI-assisted development. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI tools provide substantial assistance by suggesting solutions, generating exploratory code, identifying potential issues, and accelerating literature review and problem scoping. These tools significantly boost researcher productivity when used in a human-directed loop, though the researcher must validate and integrate outputs. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI tools substantially boost researcher productivity via code generation, debugging, literature synthesis, and idea exploration, while humans retain responsibility for problem definition and validation. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist in analyzing certain technical problems and generate code or design suggestions, the task requires deep understanding of complex system constraints, creativity in solution design, and judgment about trade-offs that current AI systems cannot reliably perform end-to-end. AI today lacks the sustained reasoning and domain depth needed to independently develop solutions for novel hardware/software problems at equal quality. |
| Task automatability | claude-sonnet-5 | 2/5 | This is a broad, open-ended R&D task requiring novel problem framing and creative synthesis across hardware/software domains, which current AI can assist but not perform end-to-end reliably.assistant |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Organizational and professional barriers are substantial: companies typically require human researchers to own critical technical decisions, intellectual property is tied to human inventors, and liability for system failures falls on the responsible engineer. Clients and stakeholders expect human expert judgment on architectural decisions. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No formal licensing requirement, but organizational trust, correctness verification for novel systems, and the need for accountable expert judgment create moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | While inference costs are low, the overhead of integrating AI outputs, validating solutions, and correcting errors for complex problems approaches or exceeds the cost of expert human analysis. The need for skilled oversight negates much of the cost advantage. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Given the task still requires substantial expert oversight and iteration, AI reduces some effort but doesn't yet replace the high-skill labor cost of a research scientist for full problem-solving cycles. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Current AI systems (GitHub Copilot, ChatGPT) can help with code generation and problem decomposition, but no deployed product reliably solves complex hardware/software problems independently. Existing tools require significant human direction, debugging, and validation; they excel at code completion rather than problem analysis and solution architecture. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI coding assistants and research tools help with subcomponents (code generation, literature review, debugging) but no deployed product autonomously analyzes and solves novel computer science research problems in production. |
Evaluate project plans and proposals to assess feasibility issues.
34CI 30–37 · exposure 30 · augmentation 75 · importance 3.5/5 · click for rater detail
Evaluate project plans and proposals to assess feasibility issues.
34| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Most research organizations still rely on human peer review and PI judgment for proposal evaluation. Adoption of AI-driven feasibility assessment is nascent; pilots exist but are not standard practice, and organizations remain cautious about automating decisions with high downstream consequences. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | R&D and tech sectors are adopting AI tools for drafting and analysis at a moderate pace, though feasibility evaluation specifically remains largely human-driven. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can significantly assist by rapidly summarizing technical risk, flagging missing requirements, and organizing proposal analysis, allowing researchers to focus on strategic and organizational judgment. This augmentation substantially raises productivity while the human researcher retains decision authority. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can efficiently summarize proposals, identify risks, check technical claims, and surface relevant literature, meaningfully speeding up human evaluators' review process. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | AI can analyze technical aspects of project plans (feasibility of algorithms, architecture) and flag obvious risks, but evaluating feasibility requires judgment about team capability, organizational constraints, market timing, and implementation interdependencies that resist full automation. A human researcher must still make the final feasibility determination. |
| Task automatability | claude-sonnet-5 | 2/5 | Assessing feasibility involves deep technical judgment, organizational context, and risk assessment that current AI can support but not reliably perform end-to-end at expert quality.:contentReference[oaicite:0]{index=0} |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Project evaluation decisions typically require sign-off from experienced researchers or managers who bear responsibility for outcomes. Organizational trust in automated feasibility judgments is limited, and there is preference for human expertise in high-stakes proposal decisions, creating moderate adoption friction. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing requirement, but organizational trust and accountability for high-stakes technical decisions create moderate friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Running AI analysis on proposals is cheap, but human oversight of AI recommendations is still necessary, and the loaded cost of a senior researcher conducting feasibility review remains substantial. The all-in cost is roughly comparable to or slightly cheaper than unaugmented human review. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Human expert review still requires substantial oversight to catch errors or missed context, so AI assistance reduces but doesn't eliminate the cost of expert time. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Tools like document analysis systems and static code review agents exist and can highlight technical issues in proposals, but no mature product reliably performs the full scope of feasibility evaluation across technical, organizational, and strategic dimensions. Deployment requires significant oversight. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI tools can summarize proposals and flag risks, but no deployed product independently evaluates research project feasibility with reliable, expert-level judgment in production. |
Maintain network hardware and software, direct network security measures, and monitor networks to ensure availability to system users.
32CI 28–37 · exposure 30 · augmentation 75 · importance 3.7/5 · click for rater detail
Maintain network hardware and software, direct network security measures, and monitor networks to ensure availability to system users.
32| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | AI-assisted monitoring tools are increasingly deployed in IT and information services, but autonomous direction of security measures and network maintenance remains limited to larger organizations with mature practices. Adoption is growing but not yet at the velocity of lower-barrier automation. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | IT/network operations sectors show moderate AI adoption via AIOps and security automation pilots, but full autonomous network management remains uncommon in production at most organizations. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI greatly enhances the productivity of network researchers through continuous monitoring dashboards, threat detection, log analysis, and anomaly flagging that surface issues for human experts to investigate and act on. This augmentation is substantial while humans remain in control of policy and critical decisions. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI-driven monitoring, anomaly detection, and automated alerting significantly boost the productivity of network administrators and security teams while humans retain oversight and decision authority. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with routine monitoring and alerting, the task requires dynamic response to novel security threats, physical hardware intervention, and judgment-based decision-making that current systems cannot fully automate end-to-end. Directing security measures and responding to complex incidents remain heavily human-dependent. |
| Task automatability | claude-sonnet-5 | 2/5 | This is a hands-on, hybrid physical/administrative task involving hardware maintenance, security policy direction, and continuous monitoring; AI can assist monitoring and anomaly detection but cannot end-to-end maintain hardware or fully direct security strategy today.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Network security has high liability and regulatory requirements (HIPAA, PCI-DSS, SOC 2); decisions about access control, threat response, and policy direction must often be signed off by authorized personnel. Security incidents have severe consequences, creating strong accountability barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No strict licensing requirement, but liability for security breaches, compliance requirements, and organizational risk aversion around network availability create meaningful friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI monitoring tools add cost through infrastructure, integration, and require specialist human oversight to validate alerts and make security decisions. The loaded cost of a research scientist remains substantially higher than AI for monitoring alone, but the combination is not order-of-magnitude cheaper. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI monitoring tools reduce some labor costs but require substantial licensing, integration, and human oversight for security decisions and physical hardware work, keeping overall costs closer to comparable rather than dramatically cheaper. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Products exist for network monitoring and anomaly detection but rely on human expertise to direct security policy and respond to novel threats. Deployed systems flag issues reliably but cannot autonomously execute complex security decisions or maintain the infrastructure without human oversight. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Products like network monitoring/AIOps tools and SIEM/security automation platforms are deployed in production for anomaly detection and alerting, but full network administration and hardware upkeep still require human network engineers. |
Design computers and the software that runs them.
32CI 25–40 · exposure 25 · augmentation 88 · importance 3.6/5 · click for rater detail
Design computers and the software that runs them.
32| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Adoption in research and development settings remains limited to narrower code-generation assistants and prototyping aids. Full system and software design automation has not achieved production-scale displacement in computer science organizations; adoption is pilot-stage and heavily human-driven. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Software and computer science R&D sectors are among the fastest adopters of AI coding and design assistance tools, with widespread production use of copilots and code-generation systems. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI significantly augments computer and software design by accelerating code synthesis, suggesting architectures, automating boilerplate, and exploring design alternatives. When kept in an assistive loop with human scientists, these tools measurably raise productivity and creativity on design tasks. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI coding assistants, architecture brainstorming tools, and code generation significantly boost productivity for researchers and engineers designing systems, while humans retain final design authority. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Current AI can assist with components of software design (code generation, architecture suggestions) but cannot end-to-end replace the full creative, strategic task of designing both computer systems and their software from requirements to deployment with 50% time savings at equal quality. The task requires deep system-level tradeoffs, novel problem-solving, and accountability that AI cannot yet autonomously handle. |
| Task automatability | claude-sonnet-5 | 2/5 | High-level architecture and software design require novel synthesis, tradeoff judgment, and organizational context that current AI cannot fully replace end-to-end, though AI can accelerate sub-components like code generation or documentation. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Significant barriers exist: design outcomes carry liability and quality risk, organizational culture values expert judgment and novelty, regulatory and contractual requirements often demand human accountability for critical systems, and cross-functional human collaboration is embedded in design workflows. These factors slow straightforward substitution. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement mandates a human for this task, but organizational trust, IP/liability concerns, and the need for deep domain judgment create moderate friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI inference costs for design tasks (multiple iterations, long context windows, human review cycles) remain comparable to or exceed the loaded cost of junior/mid-level scientists for many design problems, especially those requiring novel or high-stakes solutions. Setup and oversight costs add friction. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI tools reduce some drafting/coding costs but the overall design task still requires substantial expert human oversight, keeping all-in costs closer to human-comparable than order-of-magnitude cheaper. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | AI tools like GitHub Copilot and ChatGPT demonstrate limited feasibility: they generate code snippets and explain concepts but do not reliably perform full system or software design in production environments. Deployed products lack the end-to-end design capability, verification discipline, and accountability needed for safety-critical or complex systems. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Coding assistants and design-support tools exist and are used in production, but no deployed product autonomously designs computer systems or full software architectures reliably. |
Conduct logical analyses of business, scientific, engineering, and other technical problems, formulating mathematical models of problems for solution by computers.
31CI 25–38 · exposure 25 · augmentation 75 · importance 3.6/5 · click for rater detail
Conduct logical analyses of business, scientific, engineering, and other technical problems, formulating mathematical models of problems for solution by computers.
31| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Research and technical organizations are early-stage in adopting AI for core research problem formulation; most use it for auxiliary tasks (literature search, code generation) rather than autonomous problem modeling. Deep adoption in production research workflows remains limited. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Research and tech-adjacent fields are adopting AI copilots and modeling assistants at a moderate pace, though full automation of problem formulation remains largely at pilot stage. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI tools meaningfully assist by rapidly generating candidate models, suggesting mathematical formulations, checking logical consistency, and surfacing relevant prior work. A human researcher using these assistants can iterate and explore problem spaces faster than without them, while maintaining full control over final model selection and validation. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI substantially aids brainstorming, drafting equations, suggesting formulations, and exploring solution approaches, meaningfully speeding up the human's analytical process while the scientist retains control. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist in formulating certain mathematical models and analyzing structured problems, the task fundamentally requires human judgment to understand the business/technical context, validate assumptions, and make key decisions about problem formulation. Current AI systems lack the domain expertise and autonomous reasoning needed to handle the full end-to-end task without substantial human oversight and revision. |
| Task automatability | claude-sonnet-5 | 2/5 | AI can assist with formulating and drafting mathematical models but the core creative logical analysis of novel technical problems still requires substantial human expertise and judgment, especially for domain-specific nuances. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Formal barriers are moderate, but substantial organizational friction and liability concerns exist: businesses typically require human credibility and sign-off on technical problem formulations, and errors in model assumptions can cascade into costly downstream failures. Clients and regulators expect qualified humans to take responsibility for mathematical models. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No formal licensing requirement, but organizational trust, correctness-criticality, and the need for domain expertise create moderate friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Current AI tools and inference costs are non-trivial, and the task requires significant human verification, domain expertise input, and iterative refinement. The all-in cost per solved problem remains higher than employing an experienced research scientist for this specialized work. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Because human verification and iterative correction are usually required for correctness of a novel model, the all-in cost of using AI plus expert review is often comparable to or only modestly cheaper than a skilled scientist doing it directly. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Some deployed tools (symbolic math solvers, constraint programming libraries) can handle narrow subproblems, but no end-to-end product reliably conducts complete logical analysis and problem formulation across diverse technical domains. Research systems exist but production deployment for this cognitive task remains limited. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Current LLMs and tools can help draft formalizations for well-specified or textbook-like problems, but no deployed product reliably performs open-ended problem formulation across diverse scientific/engineering domains without heavy human oversight. |
Participate in multidisciplinary projects in areas such as virtual reality, human-computer interaction, or robotics.
29CI 20–38 · exposure 20 · augmentation 75 · importance 3.4/5 · click for rater detail
Participate in multidisciplinary projects in areas such as virtual reality, human-computer interaction, or robotics.
29| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While research institutions are adopting AI tools (simulators, code generators), actual displacement of research scientists in multidisciplinary projects is minimal. These are high-expertise roles where humans remain the primary decision-makers and project owners. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Research and tech sectors adopt AI tools quickly for coding, data analysis, and drafting, but full participation in interdisciplinary robotics/VR/HCI projects remains pilot-stage rather than routine. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can substantially augment researcher productivity through rapid prototyping, simulation tools, literature synthesis, code generation, and data analysis, enabling scientists to iterate faster and explore designs more thoroughly while maintaining human creativity and oversight. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI significantly augments literature review, code generation, simulation, and design iteration within such projects, meaningfully boosting researcher productivity while humans retain project direction. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | This task requires significant domain expertise, creative problem-solving, and interdisciplinary collaboration that current AI cannot fully automate. While AI can assist with literature reviews, simulation, or coding components, the core intellectual work of designing and executing novel multidisciplinary research projects requires human judgment, innovation, and real-time team coordination. |
| Task automatability | claude-sonnet-5 | 2/5 | Participation in multidisciplinary research projects requires original ideation, cross-domain synthesis, and collaborative decision-making that current AI cannot autonomously replace, though it can assist with subtasks like literature review or code prototyping. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Multidisciplinary research projects typically require human institutional authority, research ethics oversight, funding accountability, and peer-reviewed publication of results signed by human researchers. These legal and organizational requirements create strong barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement exists for this task, but organizational trust, intellectual credit, and the need for embodied/collaborative judgment create moderate friction against pure AI substitution. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI tools (code generation, simulation, analysis) can reduce some costs within these projects, but the human research scientist remains essential for design, decision-making, and integration. The all-in cost of AI assistance plus human oversight is comparable to or more expensive than the scientist alone. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Substituting a full research scientist's collaborative and creative contribution would require extensive human oversight and integration, making AI-only substitution costly relative to its narrow capability gains. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed product can autonomously participate meaningfully in multidisciplinary research projects. AI systems lack the ability to independently navigate research strategy, coordinate across teams, and make novel scientific decisions at the level required in VR, HCI, or robotics research. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI tools exist for coding assistance, literature synthesis, and simulation but no deployed product independently participates in or manages multidisciplinary research collaboration end-to-end. |
Develop performance standards, and evaluate work in light of established standards.
29CI 25–32 · exposure 25 · augmentation 50 · importance 3.1/5 · click for rater detail
Develop performance standards, and evaluate work in light of established standards.
29| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Research institutions and tech organizations are slow to deploy AI for performance evaluation of knowledge workers, preferring established peer-review and expert-committee processes; adoption remains largely at the pilot or analytical-support stage rather than replacement. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | R&D and tech sectors adopt AI tools quickly for many tasks, but management/evaluation functions like standard-setting see slower, more cautious adoption due to their judgment-heavy nature. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist by analyzing performance data, flagging anomalies, and suggesting metric frameworks, helping experts work faster; however, the final synthesis and judgment typically remain with human researchers, making this a moderate augmentation scenario. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help draft rubrics, benchmark comparisons, or summarize performance data, usefully supporting but not replacing the evaluative judgment involved. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Developing performance standards requires domain expertise, judgment about organizational goals, and stakeholder input that exceed current AI capabilities. AI can assist in data analysis and suggest metric structures, but cannot end-to-end define standards that meet organizational context with 50% time savings at equal quality. |
| Task automatability | claude-sonnet-5 | 2/5 | Setting performance standards and judging work against them requires domain expertise, organizational context, and value judgments that current AI cannot reliably originate end-to-end.assistance is limited to drafting or checklist support. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Performance standards for research scientists involve peer review, institutional governance, and expert judgment that carry reputational and career consequences; there is strong organizational and professional resistance to automating such consequential evaluations without human accountability. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No formal licensing requirement, but organizational trust, accountability for personnel/performance decisions, and expertise requirements create meaningful friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI evaluation tools exist but require substantial expert oversight and refinement to be trustworthy for research standards; the cost of integration, validation, and human review often approaches or exceeds the value of partial automation for such high-stakes judgments. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Human expert judgment is required for credible standard-setting, and AI assistance still needs significant expert oversight, so cost savings are modest rather than order-of-magnitude. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | While AI tools can perform partial evaluation tasks (e.g., code review, metric computation), no deployed product reliably develops comprehensive performance standards or evaluates complex research work end-to-end. This remains largely manual and expert-driven in practice. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | No deployed product autonomously creates and applies performance standards for research evaluation; AI tools exist for code review or metrics tracking but not holistic standard-setting and judgment. |
Approve, prepare, monitor, and adjust operational budgets.
26CI 25–28 · exposure 25 · augmentation 75 · importance 2.8/5 · click for rater detail
Approve, prepare, monitor, and adjust operational budgets.
26| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While finance and IT sectors have moderate adoption of budget forecasting and monitoring tools, actual approval workflow automation remains rare in production; organizations retain human budget gatekeepers for accountability. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Research and tech organizations increasingly use AI-assisted analytics tools for budgeting and forecasting, though full budget approval workflows remain human-driven. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can significantly augment this task by providing real-time budget variance analysis, automated forecasting, scenario modeling, and anomaly detection—allowing humans to make faster, more informed approval and adjustment decisions without removing them from the loop. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can significantly assist by generating budget forecasts, flagging anomalies, and modeling scenarios, improving efficiency for the human who retains approval authority. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Budget approval and adjustment require human judgment on strategic priorities, organizational constraints, and risk tolerance. While AI can assist with data gathering, forecasting, and preliminary analysis, the approval authority and final adjustment decisions remain firmly human responsibilities. |
| Task automatability | claude-sonnet-5 | 2/5 | Budget preparation and monitoring involve data aggregation AI can assist with, but 'approve' implies accountable judgment and authority that current AI cannot exercise end-to-end.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Budget approval typically requires explicit organizational authority, fiduciary responsibility, and often formal sign-off by a human in a designated role; liability for budget errors rests with accountable individuals, creating a strong legal barrier to full automation. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Budget approval typically requires organizational authority, fiduciary responsibility, and accountability that is tied to a specific role/title, creating strong institutional and governance barriers. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI tools for budget monitoring and forecasting have non-trivial integration and maintenance costs, and oversight by the human decision-maker is mandatory, keeping total cost per budget cycle high relative to the human time saved. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI tools can cheaply generate budget drafts or variance reports, but the approval and adjustment decisions still require human oversight, keeping overall cost savings modest. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product reliably performs budget approval and adjustment end-to-end; systems exist for budget forecasting and monitoring alerts, but they require substantial human override and manual decision-making in production settings. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Financial planning software with AI-assisted forecasting exists, but no deployed product autonomously approves or manages operational budgets for research organizations without human decision-making. |
Apply theoretical expertise and innovation to create or apply new technology, such as adapting principles for applying computers to new uses.
26CI 20–32 · exposure 16 · augmentation 75 · importance 3.9/5 · click for rater detail
Apply theoretical expertise and innovation to create or apply new technology, such as adapting principles for applying computers to new uses.
26| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While tech sectors adopt AI tools widely, adoption is confined to augmentation (code generation, simulation) rather than replacing the research scientist role itself. Independent innovation remains a human-centric process even in high-tech industries. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Computer science research is a fast-adopting, highly digitized field where AI coding and research assistants are widely piloted, though true innovation tasks see slower integration than routine tasks. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI substantially augments research scientists through literature synthesis, code generation, simulation, and rapid prototyping, materially accelerating the execution phase of research while humans retain strategic direction and novelty judgment. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI substantially assists brainstorming, literature synthesis, code prototyping, and exploring design spaces, meaningfully boosting researcher productivity while the human directs the innovative process. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Creating and applying new technology requires novel problem formulation, research direction setting, and creative synthesis of principles—tasks where current AI lacks independent agency and judgment. While AI can assist with literature review, code generation, and hypothesis testing, the core innovation and theoretical insight remain fundamentally human responsibilities that cannot be end-to-end automated at 50% time savings. |
| Task automatability | claude-sonnet-5 | 2/5 | This task involves original theoretical innovation and novel application of computing principles, which requires creative insight and judgment that current AI cannot reliably generate end-to-end; AI can assist but not replace the core innovation work. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Strong organizational, reputational, and legal barriers protect this role: innovation output is attributed to human researchers, institutional IP practices require human authorship and accountability, and stakeholders demand human judgment on research direction and novel technology viability. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement blocks AI involvement, but organizational trust in AI-generated novel scientific contributions and peer validation norms create moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI tools (coding assistants, research platforms) reduce peripheral task costs but cannot substitute for the high-value human expertise required for innovation direction. The loaded cost of a research scientist remains far higher than the marginal cost of AI augmentation, making substitution economically unfavorable. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI tools can reduce some exploratory and literature-review costs, but the actual innovative synthesis still requires expensive human expert time, so overall cost savings are limited. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed product performs autonomous technology creation or fundamental innovation. Current AI systems are tools for code writing and analysis, not for independently conceiving new technological applications or adapting established principles to genuinely novel domains. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously originates new theoretical computing paradigms or applies principles to genuinely novel domains; this remains research-stage even for advanced AI systems. |
Consult with users, management, vendors, and technicians to determine computing needs and system requirements.
26CI 20–32 · exposure 20 · augmentation 75 · importance 3.4/5 · click for rater detail
Consult with users, management, vendors, and technicians to determine computing needs and system requirements.
26| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Adoption of AI for requirement gathering is minimal in practice. Firms continue to rely on human consultants for this task because stakeholder engagement and judgment are seen as core professional services that justify human employment. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Tech and R&D sectors are fast adopters of AI tools generally, but for this specific consultative/requirements-gathering activity, adoption is mostly limited to note-taking and drafting support rather than replacing the interaction itself. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can meaningfully assist by analyzing stakeholder responses, drafting requirement summaries, identifying gaps and inconsistencies, and generating documentation—all of which can help a human consultant work faster and more thoroughly while maintaining final ownership of the consultation. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully help by transcribing/summarizing meetings, drafting requirement documents, generating clarifying questions, and organizing stakeholder input, boosting the researcher's efficiency while they remain the primary consultant. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can gather basic requirements and draft summaries of needs, the task fundamentally requires understanding organizational context, priorities, and stakeholder concerns that demand human judgment and relationship management. Current AI systems cannot reliably elicit the nuanced, implicit requirements that emerge from multi-party consultation without expert human oversight. |
| Task automatability | claude-sonnet-5 | 2/5 | This task requires synthesizing stakeholder input, negotiating trade-offs, and building trust across parties with differing priorities—AI can assist but cannot conduct these consultative relationships end-to-end today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Strong barriers exist: organizations typically require credentialed professionals to own systems requirements; liability for system failures falls on the human decision-maker; and customers and stakeholders expect direct engagement with qualified experts, not automation of this trust-dependent function. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No formal licensing requirement, but organizational trust, stakeholder relationships, and accountability for system design decisions create meaningful friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | The cost of AI tools plus integration and human oversight to validate requirement determinations would still be substantial relative to having a skilled consultant perform the consultation directly, given the critical importance of accuracy in systems design. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI can cheaply summarize meetings or draft requirement docs, but the core consultative work still requires paid human time, so overall cost savings versus a human are modest given oversight and follow-up needs. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product reliably performs this task end-to-end. Some tools assist with requirement documentation and can summarize stakeholder input, but they cannot independently conduct multi-stakeholder consultations to determine actual needs—they lack the adaptive listening and contextual understanding needed. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously runs stakeholder consultations to gather requirements; requirements-elicitation remains a human-led, interpersonal process even when AI note-taking or summarization tools are used alongside it. |
Develop and interpret organizational goals, policies, and procedures.
16CI 7–25 · exposure 13 · augmentation 63 · importance 3.2/5 · click for rater detail
Develop and interpret organizational goals, policies, and procedures.
16| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While information-sector organizations experiment with AI-assisted document drafting, actual adoption for autonomous development and interpretation of strategic goals remains minimal. This remains a human-centered decision function across virtually all organizations. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | While research/tech sectors adopt AI tools quickly for many tasks, strategic policy-setting itself sees little actual delegation to AI in practice. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can meaningfully assist by drafting policy language, analyzing competitor or regulatory frameworks, and synthesizing stakeholder feedback into documents humans review and revise. This augmentation improves efficiency without removing human responsibility for strategy. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist by summarizing data, drafting policy language, analyzing precedents, and modeling implications, boosting productivity while humans retain decision authority. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Developing organizational goals, policies, and procedures requires strategic judgment, stakeholder alignment, and organizational context that AI cannot autonomously establish. Current systems can assist with drafting language or analyzing existing frameworks, but cannot independently set organizational direction or interpret competing interests at the required level. |
| Task automatability | claude-sonnet-5 | 1/5 | This requires deep organizational context, judgment about strategic priorities, and stakeholder negotiation that AI cannot perform end-to-end today.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Organizational policy and goal-setting typically require executive authority, formal approval processes, and human accountability for strategic direction. Many governance frameworks legally or structurally require human decision-makers to own these outputs, creating strong adoption barriers. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Organizational authority, accountability, and governance structures require humans to own and be responsible for policy decisions, creating strong institutional barriers. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | The task requires domain expertise and organizational accountability that expensive human professionals provide; AI tools offer modest cost reduction through drafting assistance but cannot replace the high-value strategic labor involved in authentic goal development. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no viable AI substitute performing this function independently, so cost comparison favors the human entirely. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed products reliably perform this task end-to-end in production settings. AI tools can support document generation and policy analysis, but lack the contextual authority, stakeholder input integration, and strategic decision-making that characterize actual organizational goal-setting. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously develops or interprets organizational policy; this remains a human leadership function. |
Direct daily operations of departments, coordinating project activities with other departments.
11CI 0–21 · exposure 13 · augmentation 50 · importance 3.3/5 · click for rater detail
Direct daily operations of departments, coordinating project activities with other departments.
11| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Organizations have shown no pattern of replacing human managers with AI agents for operational direction and coordination. This remains an explicitly human leadership role, and adoption of AI management remains experimental at best; no widespread displacement is evident in production settings. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | While AI adoption in tech/R&D is fast, adoption specifically for managerial direction and interdepartmental coordination remains nascent and mostly assistive. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist managers by automating scheduling, aggregating status reports, suggesting resource rebalancing, or flagging coordination bottlenecks. These tools enhance a director's visibility and decision speed, but the human retains the core responsibility for strategy and people management. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help with scheduling, status tracking, summarizing cross-team updates, and drafting communications, aiding but not replacing the coordinating manager. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Directing daily operations and coordinating across departments involves complex judgment, stakeholder management, and real-time decision-making. Current AI can assist with scheduling and information synthesis, but cannot replace the human authority and negotiation skills required to actually direct operations or resolve departmental conflicts. |
| Task automatability | claude-sonnet-5 | 1/5 | Directing daily operations and cross-department coordination requires real-time human judgment, authority, relationship management, and organizational politics that current AI cannot perform end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Department directors typically hold formal managerial authority and accountability roles with legal/organizational responsibility. Stakeholder trust, organizational hierarchy, and contractual expectations all require a human manager to execute this task; regulatory and organizational barriers to substitution are substantial. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Managerial authority, accountability, HR/legal responsibilities, and organizational trust structures create strong barriers to full automation of this leadership function. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The cost of AI systems that might assist with parts of this task (scheduling, meeting coordination) is low, but it does not displace the manager's wage. Human oversight of daily operations remains essential and irreplaceable, so the total cost ratio remains unfavorable to full automation. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no viable AI substitute for the managerial role itself, so cost comparison favors the human wholesale; AI tools add cost as aids rather than replacements. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No mature product reliably performs operational direction and cross-departmental coordination end-to-end. Scheduling tools and communication platforms exist, but they handle logistics only; the strategic oversight and conflict resolution remain firmly human. Deployed systems cannot legally or functionally assume this role. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously runs departmental operations or manages cross-team coordination as a manager; existing tools only support scheduling or communication subtasks. |
Meet with managers, vendors, and others to solicit cooperation and resolve problems.
6CI 5–7 · exposure 0 · augmentation 50 · importance 3.6/5 · click for rater detail
Meet with managers, vendors, and others to solicit cooperation and resolve problems.
6| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Organizations have shown no meaningful adoption of AI systems to replace or conduct stakeholder meetings. The need for human judgment, relationship continuity, and personal accountability makes this task resistant to automation in practice. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | While AI note-takers and meeting assistants are spreading in professional settings, actual delegation of relationship negotiation and problem resolution to AI is minimal and slow. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI could provide minor assistance such as drafting talking points, summarizing prior conversations, or organizing meeting logistics, but these are peripheral to the core task of actually meeting and resolving issues through human negotiation. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist with meeting prep, summarization, follow-up drafting, and tracking action items, boosting the human's effectiveness in these interactions. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Meeting with managers, vendors, and others to solicit cooperation and resolve problems is fundamentally a human interaction task requiring negotiation, relationship-building, and contextual judgment. Current AI systems cannot independently conduct meetings or establish the trust-based relationships necessary for this interpersonal work. |
| Task automatability | claude-sonnet-5 | 1/5 | This is a live interpersonal negotiation and relationship-management task requiring real-time judgment, trust-building, and political navigation that current AI cannot perform end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Strong barriers exist: stakeholders expect to engage with an actual accountable person, not a system; managers and vendors require human authority and judgment; and there is an implicit human-contact requirement for genuine cooperation and problem-solving in professional relationships. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Trust, authority, and accountability in vendor/manager relationships strongly favor human presence; organizational norms and liability for commitments made in these meetings create high friction against substitution. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The cost of AI systems attempting to participate in or simulate stakeholder meetings, combined with required human oversight and the potential reputational damage from miscommunication, would far exceed the cost of a human research scientist conducting these meetings directly. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no viable AI substitute performing this task, so cost comparison favors the human by default since AI cannot deliver the output at all. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed AI product can autonomously conduct meetings to solicit cooperation and resolve problems with external stakeholders. While AI chatbots exist, they cannot replace the social presence, accountability, and decision-making authority required in professional stakeholder engagement. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously meets with human stakeholders to negotiate cooperation and resolve interpersonal/organizational problems; this remains firmly human-only in practice. |
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.