Teaching Assistants, Postsecondary
25-9044.00Median wage $42,910/yr164,090 employed (US)Rank #206 of 923 scored · top 22% by substitution
Assist faculty or other instructional staff in postsecondary institutions by performing instructional support activities, such as developing teaching materials, leading discussion groups, preparing and giving examinations, and grading examinations or papers.
Sub-scores
0–100 · band = confidence interval from rater disagreement
Substitution38
Exposure34
Augmentation61
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
20 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
10%
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%33
panel mean rating 2.3/5 → substitution pressure 33/100
Technical feasibility todayw 20%35
panel mean rating 2.4/5 → substitution pressure 35/100
Cost vs. human wagew 15%41
panel mean rating 2.7/5 → substitution pressure 41/100
Adoption barriersw 20%inverted — strong barriers lower the score50
panel mean rating 3.0/5 (barrier strength) → substitution pressure 50/100
Sector adoption velocityw 10%34
panel mean rating 2.4/5 → substitution pressure 34/100
Task breakdown (20 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.
Copy and distribute classroom materials.
97CI 97–97 · exposure 100 · augmentation 38 · importance 3.4/5 · click for rater detail
Copy and distribute classroom materials.
97| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Higher education has already extensively adopted digital learning management systems and print-on-demand services, with many institutions having moved substantially away from manual copying and distribution. Adoption is deep and ongoing across postsecondary institutions. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Higher education has broadly adopted LMS-based material distribution over the past decade, though some institutions and courses still rely on physical handouts. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI offers minimal assistance to a human performing this task, as it is primarily physical/mechanical copying work where productivity gains accrue mainly through full automation rather than human-AI collaboration or enhancement. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI/digital tools can automate most of this task, but the residual augmentation value for a human is limited since the task is nearly fully automatable already. |
| Task automatability | claude-haiku-4-5-20251001 | 5/5 | Copying and distributing classroom materials is a purely mechanical task involving document reproduction and logistics, which can be fully automated using print-on-demand systems, document management platforms, and robotic process automation for digital distribution. This task has minimal complexity and no decision-making requirements, making it trivially automatable with ≥50% time savings. |
| Task automatability | claude-sonnet-5 | 5/5 | Copying and distributing materials is a rote administrative task easily handled by digital tools (LMS uploads, email, print automation) with full time savings at equal or better quality. |
| Adoption barriers | claude-haiku-4-5-20251001 | 1/5 | No licensing, regulatory, or legal barriers exist to automating copying and distribution; this is purely operational logistics with no human-contact requirement or liability concerns. Institutions face minimal friction in deploying automated solutions. |
| Adoption barriers | claude-sonnet-5 | 1/5 | No licensing, liability, or regulatory barriers exist; distributing course materials digitally is already standard, unregulated practice with no requirement for human involvement. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | Automated copying and digital distribution (print-on-demand, LMS uploads) cost a fraction of the human labor required to manually copy and hand-distribute materials, representing an order-of-magnitude cost advantage for anything beyond trivial volumes. |
| Cost vs. human wage | claude-sonnet-5 | 5/5 | Digital distribution via LMS or file sharing costs a fraction of a cent per student compared to the loaded wage of a TA spending time physically copying and handing out materials. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 5/5 | Deployed products reliably perform this task at scale today: commercial printing services, learning management systems (Canvas, Blackboard), and digital distribution platforms handle document copying and distribution routinely in production educational environments. |
| Technical feasibility today | claude-sonnet-5 | 5/5 | Learning management systems (Canvas, Blackboard, Google Classroom) already reliably distribute materials digitally at scale in production today, and printing/copying automation is mature and widespread. |
Develop teaching materials, such as syllabi, visual aids, answer keys, supplementary notes, or course Web sites.
71CI 64–79 · exposure 67 · augmentation 100 · importance 4.2/5 · click for rater detail
Develop teaching materials, such as syllabi, visual aids, answer keys, supplementary notes, or course Web sites.
71| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Postsecondary institutions (information-sector, high digitization) are rapidly integrating AI into course development; faculty and TAs are openly adopting ChatGPT and similar tools for syllabus and content generation, with adoption visible in both pilots and early production use. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Higher education is adopting generative AI for content creation at a moderate pace, with growing pilot programs but inconsistent institution-wide deployment. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI dramatically accelerates TA productivity on this task—a TA can iterate, refine, and customize generated syllabi and materials far faster than writing from scratch, while retaining full pedagogical control and judgment over the final product. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially speeds up drafting of syllabi, visual aids, and supplementary content while the TA/instructor retains control over accuracy and pedagogical fit. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Current AI systems can generate substantial portions of syllabi, visual aids, answer keys, and course notes at scale with minimal human oversight; they can draft Web site content and structure. The remaining 10–20% (policy compliance, institution-specific constraints, pedagogical refinement) requires human review, easily meeting the 50% time-saving threshold. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can draft syllabi, slides, answer keys, and supplementary notes quickly, but faculty oversight, alignment with specific course goals, and accuracy checks still require substantial human editing, capping full end-to-end savings at roughly half the task. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Institution review and approval workflows create mild friction, and some faculty prefer bespoke materials; however, no legal licensing, liability shields, or regulatory mandate requires a human TA to author these materials, and many institutions explicitly encourage AI-assisted drafting. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement for creating course materials, though institutional policies, academic integrity concerns, and faculty approval create moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | A single API call or subscription tier costs pennies per syllabus or visual-aid draft, versus hours of TA labor at $15–25/hour fully loaded. Even accounting for oversight, the cost ratio favors AI by an order of magnitude or more. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Generating drafts of syllabi, slides, and notes via AI subscription costs a few dollars versus hours of TA labor, though human review time somewhat narrows the gap. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Mature products (Claude, ChatGPT, specialized tools like Canva) demonstrate reliable production-grade performance on syllabus drafting, content generation, and basic Web design. Institutions are actively deploying these in teaching support workflows; error rates on factual content require human spot-checking but are acceptable for draft-stage work. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | Deployed tools (ChatGPT, Copilot, Canva, LMS-integrated generators) are widely used by instructors and TAs today to draft these materials, though answer keys and technical content need verification. |
Evaluate and grade examinations, assignments, or papers, and record grades.
66CI 54–79 · exposure 62 · augmentation 88 · importance 4.6/5 · click for rater detail
Evaluate and grade examinations, assignments, or papers, and record grades.
66| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Higher education and testing organizations (especially large institutions with scale and digitization) are actively deploying and piloting AI-assisted grading. Adoption is measurable in production, with vendors reporting integration into learning management systems at significant institutions. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Higher education has moderate digitization and AI experimentation, with grading tools piloted widely but full-scale production use still limited by institutional caution. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI excels at augmenting human graders by pre-sorting papers, flagging potential concerns, drafting initial feedback, and applying consistent rubrics, which can double or triple grading speed while keeping instructors in the loop for final judgment and quality assurance. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI substantially speeds up grading by pre-scoring, flagging errors, and drafting feedback, letting TAs review and finalize grades faster while remaining the decision-maker. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | AI systems can now reliably grade objective components (multiple choice, true/false, short answer with rubrics) and assist substantially with essay evaluation using LLMs trained on rubric-based assessment. For standardized or rubric-heavy grading, AI achieves >50% time savings at comparable quality, though subjective nuance in open-ended work still benefits from human review. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can grade objective and structured assignments well and provide draft feedback on essays, but nuanced grading of open-ended academic work, especially at advanced levels, still requires human judgment for full reliability and equal quality. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Institutions retain final sign-off authority, but no law mandates human grading. Adoption is largely voluntary, and customer/instructor preference for human judgment provides some friction, but it is not a hard regulatory barrier to automation. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No formal licensing requirement blocks AI grading, but academic integrity policies, instructor oversight requirements, and student contestation of grades create moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | AI inference cost per assignment is negligible (cents to fractions of a cent), while a teaching assistant's loaded wage to grade one assignment is typically $5–20+. Even accounting for integration and oversight, AI is orders of magnitude cheaper per graded item at scale. |
| Cost vs. human wage | claude-sonnet-5 | 4/5 | Once set up, AI grading of assignments is very cheap per unit compared to TA hourly wages, though integration and oversight costs reduce the savings somewhat. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Multiple deployed products (Turnitin, Gradescope, Canvas plugins) and general-purpose LLMs with prompt engineering perform automated grading in production at scale, especially for objective and rubric-aligned tasks. Error rates on complex essays remain material but acceptable for oversight workflows in real institutions. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | AI grading tools (e.g., automated essay scoring, LMS-integrated graders) are deployed in some courses, but adoption is uneven and accuracy on complex or subjective assignments remains limited compared to human graders. |
Inform students of the procedures for completing and submitting class work, such as lab reports.
62CI 44–81 · exposure 55 · augmentation 75 · importance 4.1/5 · click for rater detail
Inform students of the procedures for completing and submitting class work, such as lab reports.
62| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Higher education has been slow to deploy AI agents for student-facing instructional tasks in production; most adoption remains in pilots or supplementary use (e.g., chatbots alongside TAs) rather than replacement, reflecting sector caution around student outcomes and institutional trust. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Higher education is adopting AI course tools and LMS automation steadily, but many instructors still handle such communications manually or semi-manually. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can meaningfully assist TAs by auto-generating draft instructions, creating FAQs, or drafting clarification emails that the TA reviews and personizes—substantially reducing TA time while keeping the human responsible for accuracy and student communication. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can draft, personalize, and automatically distribute submission instructions and FAQs, significantly reducing repetitive TA effort while TAs still handle edge cases. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | AI can generate initial procedural text or email templates for submission instructions, but this task involves adapting procedures to specific course contexts, responding to student questions, and ensuring comprehension—activities requiring situational judgment and interaction that are difficult to fully automate end-to-end with 50% time savings. |
| Task automatability | claude-sonnet-5 | 4/5 | This is a routine information-delivery task (explaining submission procedures) easily handled by AI via generated instructions, FAQs, or chatbots with high time savings, though initial student questions may need clarification. |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | While there are no legal licensing barriers, educational institutions often retain preference for human TAs to build rapport, answer follow-up questions, and adapt explanations on the fly; this preference and organizational inertia provide modest friction to full substitution. |
| Adoption barriers | claude-sonnet-5 | 1/5 | No licensing, liability, or regulatory barriers prevent AI from communicating administrative/procedural information to students. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 3/5 | Generating and distributing procedural instructions via AI is inexpensive relative to TA time, but integration with course management systems, oversight of accuracy, and handling edge cases make the all-in cost comparable to TA wages for this particular task. |
| Cost vs. human wage | claude-sonnet-5 | 5/5 | Generating and distributing standardized procedural text via AI costs a tiny fraction of TA time spent repeating instructions to students. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | LLMs can draft and deliver procedure documentation reliably, but deployed systems lack consistent ability to handle diverse student clarification questions, adapt to course-specific systems, or verify student understanding in the way a TA would—requiring material human oversight. |
| Technical feasibility today | claude-sonnet-5 | 4/5 | LMS-integrated chatbots and course assistants (e.g., syllabus bots, automated announcements) already deliver procedural instructions reliably in many university deployments. |
Notify instructors of errors or problems with assignments.
59CI 35–84 · exposure 50 · augmentation 63 · importance 3.1/5 · click for rater detail
Notify instructors of errors or problems with assignments.
59| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Educational institutions are rapidly adopting automated grading and plagiarism detection tools; LMS platforms (Canvas, Blackboard) now integrate AI-powered assignment analysis, and many universities have deployed such systems in production across multiple courses. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Higher education is a relatively slow-adopting sector for this kind of workflow automation, with most current AI use in course-related work concentrated on grading/content generation rather than error-flagging communications. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can significantly augment TAs by instantly flagging errors and generating detailed reports, freeing them to focus on qualitative feedback and mentoring rather than routine error-spotting. This accelerates their ability to help instructors prioritize problems. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help draft clear, well-organized notification messages and even assist in scanning assignments for inconsistencies, providing useful support even though the TA remains responsible for identifying and validating issues. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | AI can reliably identify and flag common assignment errors (syntax, formatting, logic, missing components) and notify instructors with high consistency, achieving substantial time savings. However, nuanced pedagogical issues or context-dependent problems may require human judgment, preventing a full 5 rating. |
| Task automatability | claude-sonnet-5 | 2/5 | This is a simple communication task but requires human judgment to first detect the error/problem and then decide how/when to notify the instructor, limiting full automation despite the trivial messaging component.5 |
| Adoption barriers | claude-haiku-4-5-20251001 | 2/5 | Few legal or licensing barriers exist; universities already use automated grading tools. Minor friction includes instructor preference for human review, integration with institutional systems, and occasional need for human sign-off on grade-affecting errors. |
| Adoption barriers | claude-sonnet-5 | 1/5 | No licensing, regulatory, or liability barriers prevent an AI system or TA from flagging problems and messaging instructors; it's routine internal communication. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | Automated error detection via LLMs or rule-based systems costs pennies per submission after initial setup, orders of magnitude cheaper than paying a TA to manually review and notify on each assignment. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | A human TA already performs this as a minor part of broader duties, so there's little marginal cost savings from AI since the detection work still requires human review of assignment content. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 4/5 | Deployed LLM-based systems and specialized educational software already perform automated grading and error detection in production across universities; systems reliably catch syntax errors, plagiarism, and structural issues. Minor limitations remain in interpreting ambiguous requirements or domain-specific conventions. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | No deployed product specifically monitors assignments for errors and autonomously notifies instructors; existing tools could draft a message but the detection and judgment step is not productized for this use case. |
Return assignments to students in accordance with established deadlines.
57CI 25–89 · exposure 55 · augmentation 50 · importance 4.0/5 · click for rater detail
Return assignments to students in accordance with established deadlines.
57| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Postsecondary education is a slower-adopting sector for task automation; adoption of full assignment-return automation remains rare, with most institutions relying on basic LMS scheduling rather than AI agents managing the entire process. |
| Sector adoption velocity | claude-sonnet-5 | 4/5 | Higher education has broadly adopted LMS automation for grade return and deadline management for years, making this a mature, widespread practice. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI could usefully assist by flagging overdue returns, organizing batches by deadline, or generating reminder messages, helping a TA manage the scheduling and coordination while the human retains responsibility for actual deadline adherence and exceptions. |
| Augmentation potential | claude-sonnet-5 | 3/5 | While largely automatable, AI/LMS tools still benefit from human oversight to handle exceptions like late submissions or grading disputes. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI could help organize and prepare assignments for return (batch processing, scheduling), the task requires human judgment about deadline verification, individual student circumstances, and the actual logistics of distribution. Current systems cannot fully automate the end-to-end task without substantial manual oversight. |
| Task automatability | claude-sonnet-5 | 4/5 | Returning assignments by deadline is a logistical task easily handled by LMS automation (e.g., Canvas, Gradescope auto-release schedules) with minimal human oversight needed. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Educational institutions have established deadlines, grading verification requirements, and institutional policies around assignment handling. Many institutions require faculty or official staff to manage the grading workflow, creating regulatory and organizational friction that limits substitution. |
| Adoption barriers | claude-sonnet-5 | 1/5 | No licensing, liability, or human-contact requirement attaches to the mechanical act of returning graded work on schedule. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | The task is largely administrative and low-cost for humans; integrating an AI system to handle deadline-based assignment return would require learning management system integration, monitoring, and oversight that may exceed the cost of a TA simply returning assignments on schedule. |
| Cost vs. human wage | claude-sonnet-5 | 5/5 | Automated scheduling/distribution via existing LMS software costs a tiny fraction of TA time compared to manual handling. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product reliably handles the full scope of returning assignments by deadline in academic settings. Email/LMS automation exists but requires significant manual setup and verification, and doesn't account for edge cases, makeup deadlines, or exceptions. |
| Technical feasibility today | claude-sonnet-5 | 5/5 | Learning management systems already reliably automate scheduled release/return of graded assignments at scale in real universities today. |
Tutor or mentor students who need additional instruction.
47CI 41–52 · exposure 34 · augmentation 88 · importance 3.9/5 · click for rater detail
Tutor or mentor students who need additional instruction.
47| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Educational technology adoption is accelerating; many institutions are piloting AI tutoring systems, and some large platforms (e.g., Chegg, course platforms) have integrated AI-assisted tutoring in production. This sits in the upper-middle range of sectoral digitization and deployment momentum. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Higher education is adopting AI tutoring tools moderately, with growing pilots and some scaled deployments, but adoption is uneven across institutions and disciplines. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI tutoring systems excel at augmenting human TAs by automating grading, generating personalized problem sets, identifying struggling students, and providing 24/7 availability. A TA leveraging AI tools can mentor far more students with higher-quality targeted feedback. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully augment TAs by providing supplementary practice, explanations, and homework help, freeing the TA to focus on higher-value personalized mentoring. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI tutoring systems can deliver content and practice problems at scale, they lack the adaptive responsiveness, relationship-building, and individualized diagnosis of learning gaps that human mentoring provides. Current systems struggle with the persistent, personalized support required to move students significantly ahead—well below the 50% time-saving threshold for full task replacement. |
| Task automatability | claude-sonnet-5 | 2/5 | AI tutoring tools can handle some content explanation and practice, but genuine mentoring involves relationship-building, motivation, and contextual judgment that current systems cannot fully replicate end-to-end at equal quality. |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Institutions value human mentoring for accountability and student engagement, and there are expectations (both cultural and sometimes contractual) that students receive human support. However, no strict licensing or legal barrier prevents institutions from substituting AI, though liability concerns and student retention risk create meaningful friction. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No formal licensing typically required for postsecondary TAs, but institutional norms, accreditation expectations, and student preference for human interaction create some friction against full replacement. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | Per-student inference and content delivery costs are orders of magnitude lower than paying a TA, even accounting for system development and oversight. However, integration and oversight overhead prevents the absolute lowest-cost tier (rating 5). |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI tutoring platforms are cheap per interaction, but effective mentoring requires ongoing human oversight and relationship management, keeping total cost comparable to a TA's marginal cost rather than dramatically cheaper. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | AI tutoring products exist (Squirrel AI, Carnegie Learning, etc.) and see real deployment, but they exhibit material error rates in complex reasoning, limited ability to detect misconceptions, and narrow domain coverage. They work best as supplementary tools rather than reliable end-to-end replacements for human tutoring. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Deployed AI tutoring products (e.g., Khanmigo, various adaptive learning platforms) exist and are used in production, but they are typically supplementary tools rather than full replacements for human mentoring and have narrow scope. |
Schedule and maintain regular office hours to meet with students.
45CI 11–79 · exposure 42 · augmentation 75 · importance 3.7/5 · click for rater detail
Schedule and maintain regular office hours to meet with students.
45| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 4/5 | Educational institutions and postsecondary settings have rapidly adopted automated scheduling, calendar systems, and administrative automation tools as part of broader digitization; adoption is well-established in production environments. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Higher education is a slower-adopting sector for replacing direct student interaction, though scheduling and administrative aids are gradually being adopted. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | Scheduling assistants and automated calendar management dramatically increase TA productivity by eliminating manual coordination overhead, email back-and-forth, and conflict resolution while the TA remains fully present for the actual student interactions. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help schedule appointments, send reminders, and pre-answer common questions, freeing TA time, but the substantive human interaction during office hours remains largely unaided. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | Most scheduling logistics—calendar management, conflict resolution, email reminders, and follow-up coordination—can be fully automated with calendar APIs and reminder systems that save >50% of the administrative time required; only the actual meeting presence itself remains human-required. |
| Task automatability | claude-sonnet-5 | 1/5 | Meeting with students in office hours requires real-time human interaction, mentorship, and contextual judgment that current AI cannot substitute for; the human presence is the point of the task. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | While the scheduling logistics can be automated, institutional policy, faculty/department oversight, and the inherent requirement that a human be physically or synchronously present during the actual office hours create meaningful constraints on full substitution. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing requirement per se, but institutional norms, student expectations of human contact, and pedagogical policies create meaningful friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 5/5 | Automated scheduling and reminder systems cost negligibly per task (fractions of cents per scheduled meeting) compared to the loaded wage cost of a TA manually managing calendar coordination, email, and logistics. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While scheduling tools are cheap, actually replacing the interpersonal mentoring function with AI would require expensive custom deployment and still not match human quality, so cost comparison favors humans for the core task. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 5/5 | Calendar automation, scheduling assistants (Calendly, administrative tools), and email/reminder systems are mature, widely deployed products used in production at scale across educational institutions. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product performs the role of holding office hours as a substitute for a human TA; at most AI provides scheduling tools or chatbots for basic Q&A, not the task itself. |
Order or obtain materials needed for classes.
35CI 23–47 · exposure 33 · augmentation 63 · importance 3.4/5 · click for rater detail
Order or obtain materials needed for classes.
35| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Teaching assistant roles remain in traditionally low-digitization, labor-intensive academic environments. Institutional procurement systems are legacy-heavy and change-averse; AI adoption in this specific task is minimal across the higher education sector. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Higher education administrative operations adopt AI slowly compared to finance or tech sectors, with procurement often still manual or via legacy systems. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist by drafting ordering emails, checking inventory records, or suggesting materials based on course syllabi, raising efficiency on routine parts of the task. However, the human TA must retain judgment over what is actually needed and approval authority, keeping assistance to moderate levels. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully help draft order lists, track inventory needs, and communicate with suppliers, significantly speeding up part of this task while a human finalizes and executes orders. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Ordering materials requires understanding course-specific needs, budgets, and institutional procurement rules. While AI could generate order lists or draft requests, the task involves judgment calls about what materials are needed, vendor selection, and verification—most of which still requires human oversight, limiting time savings to under 50%. |
| Task automatability | claude-sonnet-5 | 3/5 | Ordering/obtaining materials (books, supplies, lab items) involves identifying needs, sourcing vendors, and submitting requests, which AI can largely handle via list generation and procurement workflows, but physical pickup or verification often remains human.tasks.rst |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Educational institutions typically require human authorization for purchasing; PAs often must verify orders comply with departmental budgets, institutional purchasing policies, and vendor agreements. Legal and financial liability for incorrect orders creates strong friction against full automation. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement, but institutional purchasing rules, budget approval workflows, and vendor relationships create moderate organizational friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Current AI tools for procurement generate modest labor savings on email composition and list generation, but integration costs and human review/approval remain substantial. The all-in cost is unlikely to undercut the loaded wage of a teaching assistant for this narrow task. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI-assisted procurement tools could cut some administrative time cheaply, but human coordination with departments, vendors, and physical logistics still requires paid staff time, keeping costs roughly comparable. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product reliably handles the full end-to-end procurement workflow for educational institutions. While email drafting and basic inventory checking exist, the integration with institutional systems, approval chains, and vendor negotiation remains largely manual in practice. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | While AI assistants can draft purchase orders or compile supply lists, no widely deployed product autonomously manages full academic material procurement in production for TAs today. |
Prepare or proctor examinations.
32CI 32–32 · exposure 25 · augmentation 75 · importance 3.9/5 · click for rater detail
Prepare or proctor examinations.
32| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Higher education is digitizing learning platforms and experimenting with automated grading, but adoption remains uneven; many instructors still prefer human proctoring and custom exam design. Pilots are common; systematic AI proctoring and generation are not yet standard practice. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Higher education has adopted AI proctoring and exam-generation tools at moderate pace, with pilots and partial deployment common but full replacement of TAs in this role still rare. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI significantly assists TAs by generating question drafts, auto-grading objective items, and flagging suspicious submissions, raising productivity. The TA remains essential for reviewing AI suggestions, customizing exams, and handling edge cases in proctoring. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can significantly help TAs draft exam questions, create answer keys, and flag suspicious proctoring footage, meaningfully boosting efficiency while humans remain responsible for final decisions. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Generating exams and automated grading of objective questions are feasible, but proctoring requires human judgment to detect cheating, manage disruptions, and ensure test integrity. End-to-end automation with 50% time savings while maintaining equal quality is not achievable today. |
| Task automatability | claude-sonnet-5 | 2/5 | Preparing exam questions can be partially automated (drafting questions, generating variants), but proctoring requires physical/live presence, identity verification, and judgment calls that current AI cannot fully replace end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 3/5 | Institutions have academic integrity policies and instructor approval requirements for exams; proctoring involves student privacy and legal exposure around surveillance and algorithmic bias. These create moderate organizational friction but no strict legal licensing barrier. |
| Adoption barriers | claude-sonnet-5 | 3/5 | Academic integrity policies, accreditation standards, and institutional trust often require human oversight of exams, though many schools already use automated proctoring tools, creating moderate but not absolute barriers. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI tools for question generation and grading are affordable per-task, but integration with learning management systems and proctoring oversight add costs. The all-in cost per exam cycle approaches or exceeds a TA's hourly wage for the same output quality. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI-assisted exam prep can be cheap, but reliable proctoring still requires human invigilators or costly monitoring software with human review, keeping overall costs comparable to human labor. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Exam generation tools exist (question banks, AI drafting), and automated grading works for multiple-choice, but deployed proctoring systems rely on surveillance software with known limitations and high false-positive rates. No mature product reliably handles both preparation and proctoring at production scale. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI question-generation tools and online proctoring software exist, but they have material limitations (false positives in cheating detection, need for human review) and are not fully autonomous end-to-end solutions in most institutions. |
Provide instructors with assistance in the use of audiovisual equipment.
31CI 18–44 · exposure 28 · augmentation 50 · importance 3.1/5 · click for rater detail
Provide instructors with assistance in the use of audiovisual equipment.
31| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Higher education IT support for audiovisual equipment has shown slow AI adoption; institutions continue relying on teaching assistants and IT staff for in-person support. The sector prioritizes reliability and human presence for critical instructional infrastructure. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Postsecondary institutions are slow to adopt AI for physical facilities support tasks; IT/AV support remains largely human-staffed with limited AI integration beyond basic help desks. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI could assist a teaching assistant by providing step-by-step troubleshooting guides, equipment documentation lookup, or video tutorials for common problems, raising their effectiveness. However, the core manual task requires human judgment and physical action, limiting augmentation potential. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI chatbots and knowledge bases can help TAs quickly find troubleshooting steps or equipment manuals, improving efficiency for the diagnostic portion of the task even though hands-on work remains human-performed. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI could theoretically troubleshoot basic audiovisual equipment issues or provide documentation, the task inherently requires physical intervention (connecting cables, adjusting equipment, on-site presence) and real-time responsiveness to instructor needs. Current AI systems cannot reliably handle the manual, embodied aspects of equipment assistance. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can provide step-by-step guidance, troubleshooting instructions, and setup documentation for AV equipment, but physical setup, cabling, and hands-on troubleshooting still require human presence and manipulation.dd |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Educational institutions typically require staff presence on campus for real-time support, and instructors strongly prefer immediate human assistance for equipment failures. Institutional inertia, union agreements, and the expectation of human-contact problem-solving create significant adoption friction. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing or regulatory requirement exists, but the physical nature of the task (connecting cables, operating projectors, fixing hardware) creates a practical barrier to full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Even if an AI chatbot were deployed for some diagnostic guidance, the integration, maintenance, and ongoing oversight would be costly relative to a low-wage teaching assistant. Physical assistance would still require human presence, making the cost advantage minimal or negative. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI text-based troubleshooting is cheap, but since the task largely requires physical presence and manual intervention, human labor remains necessary, keeping overall cost comparable or AI-favorable only for a small info-lookup portion. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed AI product reliably provides hands-on audiovisual equipment support in educational settings. Remote troubleshooting chatbots exist but cannot replace the in-person diagnostic and hands-on assistance that instructors need during class preparation or emergencies. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Chatbots and manuals can answer basic AV troubleshooting questions, but no deployed product reliably performs the hands-on physical assistance and real-time in-room support this task typically requires. |
Provide assistance to faculty members or staff with laboratory or field research.
28CI 25–30 · exposure 25 · augmentation 63 · importance 3.8/5 · click for rater detail
Provide assistance to faculty members or staff with laboratory or field research.
28| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Academic and research institutions adopt AI tools selectively for data analysis and literature review, but remain conservative on automating core research support roles due to compliance, quality control, and mentorship requirements. Adoption is slower than in information-intensive industries. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Higher education and research settings adopt AI unevenly and cautiously for hands-on research support, with pilots for data tools but little uptake for physical assistance tasks. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can significantly augment research assistants by automating literature searches, data processing, statistical analysis, and lab documentation, allowing TAs to focus on experimental design, hands-on work, and learning. This augmentation is already emerging in computational research contexts. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can meaningfully assist with data logging, analysis, literature searches, and report drafting that accompany lab/field work, improving productivity even though physical tasks remain human-driven. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with literature searches, data organization, and computational analysis, most laboratory/field research assistance requires hands-on experimental work, equipment operation, specimen handling, and real-time troubleshooting that current systems cannot perform autonomously. Only narrow preparatory or analytical subtasks meet the 50% time-saving threshold. |
| Task automatability | claude-sonnet-5 | 2/5 | Lab/field research assistance involves physical manipulation of equipment/specimens, real-time troubleshooting, and site-specific judgment that current AI cannot perform end-to-end; only data-related sub-components (analysis, literature review) are automatable. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Faculty retain direct responsibility for research integrity, data validity, and trainee supervision; regulatory oversight of research (IRB, biosafety) mandates human accountability. Institutional research culture and the requirement for human judgment in experimental design and execution create strong adoption friction. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing requirement typically, but safety protocols, institutional research oversight, and physical presence requirements create meaningful friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Integration of AI for research assistance (specialized models, domain training, human oversight) remains costly relative to graduate/undergraduate TA wages in many fields. Current solutions are not yet an order of magnitude cheaper than human research support on a per-task basis. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | For the physical and logistical portions of the task, AI cannot substitute at all, so human labor remains necessary; only narrow analytical sub-tasks show cost savings. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Deployed AI tools can support research workflows (literature mining, statistical analysis, lab notebook assistance), but no end-to-end product reliably performs the core physical and collaborative aspects of research assistance in production environments. Existing systems have narrow scope and material limitations in novel experimental contexts. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI tools exist for data analysis, literature summarization, and protocol drafting, but no deployed product performs actual laboratory or field assistance duties reliably in production. |
Meet with supervisors to discuss students' grades or to complete required grade-related paperwork.
26CI 23–30 · exposure 25 · augmentation 50 · importance 3.7/5 · click for rater detail
Meet with supervisors to discuss students' grades or to complete required grade-related paperwork.
26| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Postsecondary education remains a laggard sector for AI automation, with strong institutional conservatism around personnel decisions, grades, and student-facing interactions; adoption of AI for supervisory meetings is minimal. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Higher education administrative processes adopt AI slowly, with pilots mainly in feedback/grading assistance rather than the interpersonal supervisory meeting itself. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist by drafting grade summaries, organizing student performance data, and auto-generating paperwork templates that a TA reviews and modifies before the meeting, moderately increasing preparation efficiency. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help organize grade data, draft paperwork, and summarize student performance trends beforehand, meaningfully aiding preparation for the meeting even if not the discussion itself. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Grade data can be automatically compiled and paperwork forms can be prefilled, but the supervisory discussion requires contextual judgment about student performance, intervention strategies, and interpersonal communication that AI cannot reliably handle end-to-end today. |
| Task automatability | claude-sonnet-5 | 2/5 | This task involves real-time interpersonal meetings with supervisors and judgment-based discussion of student performance; AI can support prep but cannot conduct the meeting or make the professor-level judgment calls itself.rn |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Academic institutions have strong norms and policies requiring human supervisors to make grade judgments and conduct face-to-face feedback meetings; there is implicit requirement for human accountability and professional relationship-building in these discussions. |
| Adoption barriers | claude-sonnet-5 | 3/5 | While not licensed activity, grading and academic records carry institutional policies, FERPA-related privacy concerns, and expectations of human oversight and accountability for grade decisions. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI can assist with data organization and form completion at near-zero marginal cost, but the meeting itself still requires a human TA, so total labor displacement is minimal; the human cost dominates. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Since the core task requires human presence and interaction, AI cannot substitute for the labor cost of the meeting itself, only marginally reduce paperwork prep time. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Systems can draft grade summaries and auto-populate forms, but no deployed product reliably conducts autonomous supervisory meetings or handles the nuanced discussion of student progress at the quality expected in academic settings. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | No deployed product autonomously conducts supervisor meetings or grade discussions; at best AI tools assist with drafting grade summaries or paperwork templates beforehand. |
Complete laboratory projects prior to assigning them to students so that any needed modifications can be made.
25CI 20–30 · exposure 20 · augmentation 50 · importance 3.8/5 · click for rater detail
Complete laboratory projects prior to assigning them to students so that any needed modifications can be made.
25| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Educational institutions adopt AI slowly and cautiously, particularly in laboratory contexts where safety and hands-on learning are core values. Current adoption of AI in TA laboratory preparation tasks remains limited to planning assistance, not execution. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Higher education and lab-based instruction are slower AI adopters for hands-on tasks compared to purely digital/information work, with pilots for content generation but not physical lab testing. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can meaningfully assist TAs by generating experimental designs, predicting common errors, and drafting procedures, improving planning efficiency and completeness. However, the physical execution of the lab work still requires human presence and judgment. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help draft lab instructions, predict expected outcomes, identify likely ambiguities or errors in written procedures, and suggest modifications, meaningfully aiding the human's review process. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can help design experimental procedures and identify potential issues, completing full laboratory projects requires hands-on physical execution, material manipulation, and real-time troubleshooting that current AI systems cannot perform end-to-end. AI could assist in planning and documentation, but cannot replace the actual lab work. |
| Task automatability | claude-sonnet-5 | 2/5 | This requires physical or applied execution of a lab procedure (or discipline-specific problem-solving) to catch practical issues, which current general AI cannot perform end-to-end; AI can assist with parts like reviewing instructions or predicting theoretical results but not physically run experiments or catch real hardware/material issues. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Educational institutions have strong liability, safety, and regulatory requirements around laboratory work. Lab completion typically requires direct human responsibility and sign-off for safety and compliance reasons, creating substantial legal and institutional barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing barrier per se, but physical lab safety, equipment access, and institutional requirements for hands-on verification create practical friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI assistance for planning and documentation is cheap, but the core task (performing the lab work) still requires a human at comparable or lower cost than AI oversight would provide, especially given the specialized equipment and safety considerations involved. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Since AI cannot substitute for physically running the lab project, the human cost is unavoidable; any AI assistance (e.g., reviewing instructions) adds marginal savings but does not meaningfully change the cost equation. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product can reliably execute complete laboratory projects independently. AI tools can draft experimental plans and identify procedural problems, but actual lab execution—mixing chemicals, operating equipment, observing results—remains human-dependent. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously completes and validates laboratory exercises across disciplines; this remains firmly in the domain of human hands-on testing, especially for wet-lab or equipment-based sciences. |
Lead discussion sections, tutorials, or laboratory sections.
18CI 5–30 · exposure 13 · augmentation 63 · importance 4.3/5 · click for rater detail
Lead discussion sections, tutorials, or laboratory sections.
18| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Adoption of AI as a discussion leader in postsecondary education remains extremely limited; institutions are cautious about delegating direct student instruction to AI. Current trend is toward AI as a supplementary tool, not a replacement for discussion leaders. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Higher education adopts AI tools for grading and content support at a moderate pace, but live instructional delivery remains a laggard area with mostly pilot-stage experimentation. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist TAs by generating discussion prompts, summarizing student contributions, drafting feedback, or providing background research, which could improve session quality and TA efficiency. However, the core task—live facilitation—remains human-led, so augmentation is partial. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully support TAs by generating discussion prompts, summarizing readings, drafting quiz questions, and providing on-demand supplementary explanations, enhancing preparation and follow-up even though it doesn't replace live facilitation. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Leading discussion sections requires real-time interactive facilitation, dynamic pedagogical judgment, and responsive engagement with diverse student questions—tasks that current AI systems cannot perform end-to-end reliably. AI cannot substitute for the human presence, authority, and adaptability needed to manage classroom discourse and assess student understanding in situ. |
| Task automatability | claude-sonnet-5 | 2/5 | Leading live discussions, tutorials, or labs requires real-time facilitation, adapting to student questions, and managing group dynamics, which current AI cannot fully replicate in-person or synchronously.dressing individual needs on the fly is beyond current automation. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Educational institutions have strong liability concerns, accreditation requirements, and student-contact norms that prefer human facilitation. Many institutions legally require qualified human instructional staff to lead sections; institutional inertia and faculty/student resistance to non-human leaders in active pedagogy are substantial. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No strict licensing requirement for TAs, but strong institutional norms, accreditation expectations, and student preference for human instructors create moderate friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | Running an AI system to lead live sessions (inference, moderation, monitoring, fallback human oversight) remains more expensive than employing a teaching assistant, particularly when accounting for liability and quality assurance in educational contexts. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI chatbots are cheap per interaction, replicating the full function of leading a live section would require robotics/multimodal presence and human oversight, making all-in cost comparable or higher than a TA's wage for equivalent quality. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | While AI can generate discussion prompts or provide tutoring-like responses asynchronously, no deployed product reliably leads live discussion sections. Prototypes and research systems exist, but production use of AI as a discussion leader at scale is negligible, and error tolerance in pedagogical settings remains low. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously leads a classroom discussion or lab section today; AI tutoring tools exist but operate in narrow, individualized, asynchronous contexts, not as substitutes for live group facilitation. |
Teach undergraduate-level courses.
16CI 7–25 · exposure 17 · augmentation 63 · importance 4.6/5 · click for rater detail
Teach undergraduate-level courses.
16| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Higher education is a laggard sector for autonomous AI adoption; most pilots remain in supplementary tutoring or grading support, not primary instruction. Production-level displacement of teaching roles is minimal, and cultural/regulatory headwinds slow adoption. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Higher education is a relatively slow-adopting sector for full instructional automation, with pilots in AI tutoring but production replacement of teaching itself remains rare. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI tools can assist teaching assistants by drafting responses, generating rubrics, automating grading, and providing learning analytics, improving productivity on administrative and preparation tasks. However, the core interactive and mentoring aspects of teaching remain largely human-driven. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI significantly aids lecture prep, slide creation, grading support, and answering routine student questions, boosting instructor productivity while they remain in charge of teaching. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Teaching undergraduate-level courses requires real-time interaction, dynamic adaptation to student questions, relationship-building, and judgment calls about pedagogical approach—none of which current AI systems can handle end-to-end without continuous human oversight. AI cannot legally or credibly replace the instructor role. |
| Task automatability | claude-sonnet-5 | 2/5 | Teaching a live undergraduate course involves real-time facilitation, adapting to student questions, and building rapport—AI can generate content but cannot fully replace the interactive, in-person teaching role at equal quality yet. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Strong barriers exist: institutions require credentialed instructors to be accountable, accreditation bodies mandate human instruction, and legal/liability frameworks place responsibility on named humans. Substituting AI for teaching requires institutional policy change and carries reputational risk. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Accreditation, institutional policy, and student expectations generally require a credentialed human instructor of record, creating strong organizational and quasi-regulatory barriers. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The all-in cost of maintaining a reliable AI teaching system (inference, content curation, oversight, liability insurance, institutional integration) currently exceeds the loaded wage of a teaching assistant, especially when accounting for the quality bar and regulatory/institutional demands. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI content generation is cheap, delivering an actual course still requires substantial human oversight, live interaction, and institutional integration, keeping all-in costs closer to human-comparable. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | While AI tutoring systems and content generation tools exist in research and limited deployments, no mature product reliably teaches full undergraduate courses at scale. Deployed systems are narrow (single-topic tutors), have material error rates, and cannot handle the breadth and adaptability required of a teaching assistant. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI tutoring tools and chatbots exist but no deployed product independently teaches a full undergraduate course in place of a human instructor at scale. |
Assist faculty members or staff with student conferences.
15CI 5–25 · exposure 8 · augmentation 38 · importance 3.0/5 · click for rater detail
Assist faculty members or staff with student conferences.
15| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Educational institutions remain slow to adopt AI for direct student-facing roles, with particular caution around student conferences where human judgment and interpersonal connection are valued; adoption of AI to replace this task is minimal despite digital expansion elsewhere in higher ed. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Higher education is a moderate-to-slow adopter of AI for interpersonal advising tasks, with pilots for scheduling or chatbots but little penetration into actual conference support. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI could assist by pre-generating student summaries, flagging academic issues, or drafting follow-up notes, but the live conference interaction itself offers limited augmentation since the teaching assistant's primary value is their direct engagement with the student. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help prepare materials, summarize student records, draft agendas, or take notes during conferences, providing useful but partial productivity support. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI could help prepare materials or summarize student performance before a conference, the core task—being present during and actively assisting in a student conference—requires real-time human judgment, interpersonal presence, and responsiveness that current AI systems cannot replicate end-to-end. |
| Task automatability | claude-sonnet-5 | 1/5 | Assisting with student conferences requires real-time interpersonal presence, scheduling coordination, and contextual judgment about student needs that current AI cannot perform end-to-end.- |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Strong barriers exist: educational institutions prioritize human interaction for pastoral and developmental purposes, students expect human presence, and there are implicit institutional norms around human-conducted advising; some institutions may have policies requiring human staff in student-facing roles. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing requirement bars this, but strong human-contact expectations and institutional norms around personal mentorship and advising create meaningful friction against substitution. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The loaded cost of an AI system capable of attending and meaningfully assisting in conferences (with required integration, oversight, and error handling) would exceed the wage of an actual teaching assistant for this specific activity. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Human TAs are relatively low-cost already, and AI cannot substitute for the interpersonal role, so any AI use is additive (scheduling/notes) rather than a cost-saving replacement. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed product reliably performs this task autonomously in production; there are no demonstrations of AI systems effectively assisting in live student conferences without human guidance and intervention. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously assists faculty during student conferences; at most AI tools help with scheduling or note-taking as peripheral support, not the core task. |
Arrange for supervisors to conduct teaching observations and provide feedback about teaching performance.
13CI 5–21 · exposure 5 · augmentation 38 · importance 3.3/5 · click for rater detail
Arrange for supervisors to conduct teaching observations and provide feedback about teaching performance.
13| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Educational institutions adopt AI slowly and cautiously, especially for tasks involving faculty supervision and performance feedback. Postsecondary teaching environments show low automation velocity for coordination and management tasks. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Higher education administrative processes are slow to adopt AI tools relative to information/finance sectors, with scheduling automation only lightly used for such tasks. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI could assist marginally by drafting emails, suggesting observation dates, or maintaining scheduling records, but the interpersonal and institutional aspects of arranging feedback make meaningful augmentation limited. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI scheduling tools and calendar assistants can help coordinate observation times and reminders, providing moderate productivity assistance while humans still lead the substantive process. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | This task fundamentally requires human judgment about scheduling, coordination with supervisors, and managing interpersonal dynamics. While AI could draft scheduling emails or calendar invitations, the core task—arranging observations and feedback—requires institutional authority and relationship management that AI cannot execute. |
| Task automatability | claude-sonnet-5 | 1/5 | This is a coordination and scheduling task tied to human judgment-based feedback on teaching quality; AI cannot conduct or arrange the substantive supervisory observation and feedback process itself., though it could help schedule meetings. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Strong barriers exist: supervisors must independently decide to conduct observations and provide authentic feedback; institutional policies often govern observation procedures; human judgment about teaching performance is legally and ethically non-delegable to machines. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No formal licensing requirement, but institutional norms, faculty governance, and human relational elements in teaching evaluation create moderate organizational friction against full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | A teaching assistant's time arranging observations is paid at modest wages (~$20–30/hour). The cost of AI systems plus oversight to attempt this coordination would exceed the direct labor cost, offering no economic advantage. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Any AI cost savings are limited to trivial scheduling automation; the core task requires human supervisors and administrative coordination that still incurs comparable labor costs. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed AI systems can autonomously arrange supervisor observations or coordinate institutional feedback processes today. This requires access to institutional scheduling systems, authority relationships, and human-to-human coordination that current AI tools cannot handle independently. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Scheduling assistants exist and could handle calendar coordination, but no deployed product manages the full arrangement of observation logistics plus feedback delivery in academic settings. |
Demonstrate use of laboratory equipment and enforce laboratory rules.
9CI 5–14 · exposure 8 · augmentation 38 · importance 3.8/5 · click for rater detail
Demonstrate use of laboratory equipment and enforce laboratory rules.
9| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Postsecondary lab instruction is slow to adopt automation in this specific task; institutions maintain human teaching assistants for safety, accreditation, and hands-on mentoring, and there is no meaningful trend of replacement with AI systems in production. |
| Sector adoption velocity | claude-sonnet-5 | 1/5 | Postsecondary lab instruction is a low-digitization, physically-grounded educational context with minimal AI displacement of hands-on supervisory roles. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can augment teaching assistants by generating instructional materials, creating visual aids, drafting lab safety documentation, and even analyzing common student errors—but the human remains essential for live demonstration and enforcement, so augmentation is real but partial. |
| Augmentation potential | claude-sonnet-5 | 2/5 | AI could help prepare instructional materials or safety checklists beforehand, but offers little real-time assistance during physical demonstration and rule enforcement. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI could assist with creating instructional videos or written guides for lab equipment use, the physical demonstration itself and real-time enforcement of rules in a lab setting require human presence and embodied judgment that current AI systems cannot replicate in real-time with sufficient quality. |
| Task automatability | claude-sonnet-5 | 1/5 | This requires physical presence, hands-on demonstration of equipment, and real-time enforcement of safety rules in a physical lab space—capabilities current AI systems lack entirely. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Strong barriers exist: institutional liability for lab safety falls on credentialed humans, many labs have explicit regulatory or insurance requirements that a human instructor must be present and demonstrate procedures, and student safety concerns create high error-cost asymmetry. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Lab safety enforcement often carries institutional liability and safety-regulation requirements, plus the need for real-time human judgment and authority, making substitution highly constrained. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | A teaching assistant's core value in this task lies in their physical presence, embodied knowledge, and real-time judgment; the all-in cost of AI systems (including robots, sensors, oversight, and liability) would far exceed a teaching assistant's loaded wage for this work. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no viable AI substitute for this physical, supervisory task, so cost comparison favors the human by default. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed product reliably demonstrates lab equipment or enforces laboratory rules in situ. Instructional videos and simulations exist but do not constitute genuine demonstration or rule enforcement with human accountability and presence. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product can physically demonstrate lab equipment use or enforce in-person safety compliance; this remains firmly in the physical/human domain. |
Attend lectures given by the supervising instructor.
6CI 0–13 · exposure 0 · augmentation 25 · importance 3.4/5 · click for rater detail
Attend lectures given by the supervising instructor.
6| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | This task involves mandatory physical or synchronous virtual presence, which operates in regulatory and organizational contexts where substitution is not permitted or meaningful. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Higher education is adopting AI tools for administrative and grading tasks but has not adopted any substitute for physically attending lectures as a TA duty. |
| Augmentation potential | claude-haiku-4-5-20251001 | 1/5 | AI cannot meaningfully augment lecture attendance itself; the value of a TA attending is the human presence, attention, and readiness to assist, which AI cannot enhance. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help TAs by providing recorded transcripts, summaries, or note-taking assistance during or after lectures, improving retention and follow-up prep even though it doesn't replace attendance. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Attending lectures is a passive presence task with no meaningful work output to automate. AI cannot substitute for physical or virtual attendance and the cognitive engagement required to maintain oversight of instruction. |
| Task automatability | claude-sonnet-5 | 1/5 | Physical or synchronous attendance at a lecture to observe, learn course content, and prepare for grading/discussion sections requires human presence and cannot be delegated to AI as an end-to-end substitute for the TA's role. |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Institutional and legal requirements mandate that TAs be present and available; many supervising instructors have explicit expectations that TAs attend lectures to support students and the course. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing barrier exists, but strong organizational/relational norms require the TA's physical or live presence for mentorship, in-class support, and instructor coordination. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | There is no cost-effective way for AI to attend lectures because the task itself has minimal direct cost—attendance is a prerequisite for other TA duties, not a billable activity. |
| 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 the AI alternative doesn't functionally replace attendance. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed product can attend a lecture in place of a human TA; this requires live presence and real-time situational awareness that current AI systems do not provide. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product 'attends' lectures on behalf of a person in a way that fulfills the institutional/relational purpose of a TA being present; transcription tools exist but don't perform the task itself. |
Related occupations — Educational Instruction & Library
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.