Paper deep dive
Who Delegates to AI? Evidence from 53,000 Agent Configurations
Hyeongjae Lee, Jihyang Cheon, Lanu Kim
Intelligence
Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 93%
Last extracted: 8/24/2026, 4:15:25 AM
Summary
This paper introduces the Agentic Adoption Index (AAI) to measure 'delegated exposure' to AI, defined as the extent to which workers have integrated AI agents into their workflows. Using 53,000 agent skill specifications from the Manus Skills Marketplace and 18,000 O*NET task statements, the authors compute semantic similarity to aggregate adoption levels by occupation. Key findings indicate that AI delegation concentrates in different occupations than previously predicted by capability-based frameworks, aligns more closely with technical feasibility than observed conversational use, and follows an inverted-U pattern across wages and education levels, peaking at the bachelor's level. The study suggests that professional discretion and work complexity, rather than just technical availability, drive adoption at the top of the labor market.
Entities (9)
Relation Signals (8)
O*NET → provides → task statements
confidence 99% · compute their semantic similarity to about 18,000 O*NET task statements...
Agentic Adoption Index → measures → delegated exposure
confidence 98% · We operationalize it as the Agentic Adoption Index (AAI), which measures how closely an occupation's tasks match the agentic routines practitioners have already built and shared.
Manus Skills Marketplace → provides → agent skill specifications
confidence 97% · We embed roughly 53,000 agent skill specifications from the Manus Skills Marketplace...
Agentic Adoption Index → peaksat → bachelor's level
confidence 95% · the AAI peaks below the top of the wage distribution and at the bachelor's level, declining at both extremes.
Agentic Adoption Index → comparedwith → Eloundou et al.
confidence 90% · The task-level exposure ratings of Eloundou et al. [2024]... capture availability exposure...
Agentic Adoption Index → comparedwith → Massenkoff and McCrory
confidence 90% · The usage-based measure of Massenkoff and McCrory [2026] captures observed exposure...
Agentic Adoption Index → comparedwith → Tomei and Teeselink
confidence 90% · The RL Feasibility Index of Tomei and Teeselink [2026] captures capability exposure...
Agentic Adoption Index → →
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:A growing literature measures how far occupations are exposed to AI, but these measures capture where AI could perform tasks, not whether workers have adopted it. We propose a new layer of exposure, delegated exposure, which records whether a worker has committed a task to AI by building it into a workflow. We operationalize it as the Agentic Adoption Index (AAI), which measures how closely an occupation's tasks match the agentic routines practitioners have already built and shared. We embed roughly 53,000 agent skill specifications from the Manus Skills Marketplace, compute their semantic similarity to about 18,000 O*NET task statements, and aggregate to the occupation level. Three findings follow. First, the occupations where delegation concentrates differ sharply from those pre-AI frameworks identified as most at risk. Second, the AAI tracks what AI could do more closely than what workers currently use it for. Third, the AAI peaks below the top of the wage distribution and at the bachelor's level, declining at both extremes. Technical availability explains most of this variation, but not the shortfall among the most educated occupations, so feasibility alone cannot account for who adopts. That shortfall may reflect work that resists advance specification, or professional discretion over the pace of codification. Distinguishing the two, and tracking how these measures diverge over time, will require repeated measurement.
Tags
Links
- Source: https://arxiv.org/abs/2608.20425v1
- Canonical: https://arxiv.org/abs/2608.20425v1
Trouble viewing inline? Open PDF directly →
Full Text
59,705 characters extracted from source content.
Expand or collapse full text
Who Delegates to AI? Evidence from 53,000 Agent Configurations Hyeongjae Lee 1 , Jihyang Cheon 2 , and Lanu Kim †1,2 1 Graduate School of Digital Humanities and Social Sciences, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Republic of Korea 2 Graduate School of Data Science, Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Republic of Korea August 24, 2026 Abstract A growing literature measures how far occupations are exposed to AI, but these mea- sures capture where AI could perform tasks, not whether workers have adopted it. We propose a new layer of exposure, delegated exposure, which records whether a worker has committed a task to AI by building it into a workflow. We operationalize it as the Agen- tic Adoption Index (AAI), which measures how closely an occupation’s tasks match the agentic routines practitioners have already built and shared. We embed roughly 53,000 agent skill specifications from the Manus Skills Marketplace, compute their semantic similarity to about 18,000 O*NET task statements, and aggregate to the occupation level. Three findings follow. First, the occupations where delegation concentrates differ sharply from those pre-AI frameworks identified as most at risk. Second, the AAI tracks what AI could do more closely than what workers currently use it for. Third, the AAI peaks below the top of the wage distribution and at the bachelor’s level, declining at both extremes. Technical availability explains most of this variation, but not the short- fall among the most educated occupations, so feasibility alone cannot account for who adopts. That shortfall may reflect work that resists advance specification, or professional discretion over the pace of codification. Distinguishing the two, and tracking how these measures diverge over time, will require repeated measurement. 1 Introduction While artificial intelligence now sits at the center of technological change, accurately measuring its labor market impacts remains an ongoing challenge. Several factors compound this difficulty: the technology is evolving rapidly, its reach is exceptionally broad—extending to unstructured tasks across diverse occupations, and we still lack a clear picture of who is adopting these tools and for what purposes [Bick et al., 2026, Frank et al., 2026]. These characteristics make AI’s labor- market impacts harder to predict than those of earlier automation [Frank et al., 2019], motivating a growing body of research that measures how exposed the tasks within occupations are to AI, and more recently, how far that exposure has translated into use. 1 arXiv:2608.20425v1 [cs.AI] 19 Aug 2026 We group the extensive literature on occupational exposure to AI or automation technology into three lines of research, ordered by how closely each comes to observing the realized adoption of AI. The first assesses which tasks AI is capable of performing, then identifies the occupations in which those capabilities would be useful [Felten et al., 2018, 2019]; what it measures is a technological ceiling. The second narrows that ceiling to what firms have chosen to build and sell: the commer- cially available applications through which large language models actually reach users [Felten et al., 2023, Eloundou et al., 2024]. The third narrows it again, to what workers choose to use, examining which work-related tasks actually appear in people’s interactions with conversational LLMs [Handa et al., 2025, Massenkoff and McCrory, 2026, Tomlinson et al., 2025]. Despite the variety of meth- ods, measures, and data they employ, these studies converge: whereas earlier automation displaced routine work, AI reaches cognitively demanding work [Frank et al., 2019, Eloundou et al., 2024, Handa et al., 2025]. These three lines share a common limit: each measures how far AI could or does touch an occupation’s work, not whether a worker has committed the work to it. We propose a fourth, which we call delegated exposure. What technology — or commercialized technology — is capable of has grown distant from what people actually adopt in practice [del Rio-Chanona et al., 2025], and the gap matters especially now that the technology is moving toward agentic systems: a recent class of LLM-based systems that pursue specified goals and carry out recurring activities through structured workflows [Yang et al., 2025, Gupta and Kumar, 2026, Xing et al., 2026]. Setting up such a system requires practitioners to define its goals, instructions, and workflow in advance, and because these configurations automate complex, multi-step work, each one records a deliberate choice to deploy AI for a specific task. That is a different kind of evidence from a conversational log: it shows not just that a worker used AI, but that they delegated a task to it. We therefore ask where delegated exposure is concentrating in the occupational structure, and how that distribution compares with what the three existing measures would predict. To measure this, we observe adoption directly, measuring how much of each occupation’s task content practitioners have already built agents for. We construct an Agentic Adoption Index (AAI) from the Manus Skills Marketplace, the one and only publicly accessible repository where practition- ers share the skill files that configure their agents. Using Sentence-BERT (all-MiniLM-L6-v2), we convert roughly 53,000 agent skill specifications into embeddings, vector representations that cap- ture their meaning. We then compute the semantic similarity between these embeddings and those of about 18,000 O*NET task statements, and aggregate the resulting similarities to the occupation level, weighting each task by its importance within the occupation. The AAI therefore captures how closely an occupation’s tasks match the agentic routines that practitioners have already built and shared, and is designed for comparison across occupations rather than interpretation on an absolute scale. Contributors are practitioners who both build agents and choose to publish them, so the measure reflects tasks whose routines generalize well enough to be worth sharing. Since the marketplace grows continuously, the same procedure can be repeated to capture shifts in what practitioners choose to automate, offering a way to track adoption as it unfolds rather than only in retrospect. Such tracking may suggest an early indicator of where AI is beginning to reshape work, to the extent that agent configurations appear before their impacts register in employment statistics. Our analysis of the AAI yields three findings. First, the occupations where agent adoption concentrates differ sharply from those that earlier automation research identified as most exposed. Second, the AAI broadly aligns more closely with measures of what AI is technically capable of than with observed patterns of conversational LLM use. Because building an agent requires deliberate configuration, practitioners invest only where the technology can reliably do the work, whereas conversational use is cheap enough to spread without regard to capability. Third, this capability- 2 driven alignment breaks down at the top. Adoption follows an inverted-U across wages and peaks at the bachelor’s level, falling off among the highest-paid and most-educated occupations. Availability accounts for the low end, where current tools cannot yet handle the work, but not the top, where occupations rank high on technical exposure and adopt agents least. Technical feasibility is therefore necessary for agentic adoption but not sufficient to explain where it occurs. This paper contributes a way to measure the adoption of agentic AI as it happens, rather than inferring it from what the technology can do — a distinction that matters because capability does not guarantee use. The measure has two practical advantages. Because it draws on a continuously growing repository, it can be recomputed as adoption evolves, unlike exposure measures anchored to periodic expert assessments. And because skill files are artifacts practitioners built for their own work, the measure captures revealed preference rather than judgments about what AI could plausibly do. Substantively, our findings speak to the sociology of professions: the occupations best positioned to automate their work are not the ones doing so, suggesting that adoption at the top of the labor market is governed by something other than technical feasibility. 2 Background and Related Work 2.1 Four layers of AI exposure Occupations are not single, uniform jobs: each comprises a distinct bundle of tasks and draws on different skills and abilities. This matters because exposure varies considerably across the tasks within a single occupation, so a measure that treats jobs as indivisible units can obscure which parts of the work are actually at stake. Research on AI’s impact at work has therefore converged on a common strategy, made possible by O*NET, a well-established dataset that specifies these elements in fine-grained form. Studies assess how far individual tasks [Eloundou et al., 2024, Massenkoff and McCrory, 2026], work activities [Brynjolfsson et al., 2018, Tomlinson et al., 2025], or skills [Felten et al., 2018, 2021, Chopra et al., 2025] will be affected by AI, then aggregate those assessments to the occupation level. We follow this strategy in our analysis. While sharing this approach, the literature diverges on which technology it measures. Mouchel et al. [Mouchel et al., 2026] and Yin and Ogut [Yin and Ogut, 2026] each distinguish capability and availability from use, but neither specifies availability from capability. We therefore organize this research into three layers of AI’s reach: what AI could technically do, what commercial tools make usable, and what people actually do with it (Fig. 1). The layers are approximately nested, since not every technical capability is built into an available tool, and not every available tool is ultimately used. We review each in turn. The first layer measures what AI could technically do, in order to identify which occupations it would affect. Brynjolfsson, Mitchell, and Rock rated individual O*NET tasks and confirmed that tasks within the same occupation often differ sharply in their susceptibility to AI [Brynjolfsson et al., 2018]. Felten, Raj, and Seamans linked AI benchmark performance to the abilities each occupation requires, later applying the same method to LLMs and finding language-heavy, cognitive occupations to be the most exposed [Felten et al., 2018, 2019, 2021, 2023]. Tomei and Teeselink extend the approach beyond language models, assessing which tasks reinforcement learning could automate [Tomei and Teeselink, 2026]. They find high exposure among monitoring and control occupations, such as gas plant operators and railroad conductors, that LLM-based measures overlook — a reminder that capability exposure depends on which technology is being measured. What these studies share is a focus on technical feasibility rather than on whether AI is used in any actual workplace. We group them under capability exposure: the extent to which a task could be affected given what AI can technically do. 3 Capability alone does not put AI in a worker’s hands. Technology can be adopted broadly only once it is built into software or an interface that workers can reach. Eloundou et al. accordingly measure not what an LLM alone could do, but how much an LLM paired with plausible software scaffolding could speed up a task, capturing potential access rather than raw capability [Eloundou et al., 2024]. Chopra et al. map over 13,000 production-ready AI tools onto occupational skills, providing a measure of where AI capabilities are available through production-ready tools across the workforce [Chopra et al., 2025]. Other studies infer availability from the supply and deployment of AI applications, using data on AI products developed by venture-funded startups [Fenoaltea et al., 2026], AI deployment events [Demirev, 2026], or administrative data on AI applications [de Souza, 2026]. Together, these findings show that what AI reaches in practice is both smaller and differently distributed than what capability measures imply. We group these under availability exposure: the extent to which a task is served by AI-enabled tools that already exist. The availability of a tool does not ensure its use; a worker must judge it useful and worth adopting. Studies of usage logs reveal this gap directly. Massenkoff and McCrory find that actual LLM use covers far less of an occupation’s activities than capability-based estimates predict, even in occupations those estimates rank as highly exposed [Massenkoff and McCrory, 2026]. Where AI is used, moreover, that use clusters in a narrow range of activities: studies of Microsoft Copilot find information work to be the dominant case [Tomlinson et al., 2025], and analysis of Claude conversations shows a similar concentration in software development and writing [Handa et al., 2025]. We group these under observed exposure: the extent to which AI is actually used for a task in practice. Observed exposure thus comes closest of the three to real adoption, yet it still does not show whether AI has become part of how work is done. A usage record establishes that AI was used, not that a worker has come to rely on it. Two features of conversational logs make this ambiguity hard to resolve. First, sessions do not align neatly with tasks: one task may span several sessions, while one session may combine several tasks [Yang et al., 2026]. Second, a prompt requires little commitment, so the same log entry could reflect casual experimentation or genuine reliance [Handa et al., 2025]. Measuring how far AI has entered work therefore requires evidence of a different kind: evidence that a worker has chosen to delegate a task by building AI into a workflow. This is what we called delegated exposure above, and develop its concept in the next section. 2.2 Measuring delegated exposure through agentic adoption Practitioner-configured agents provide clearer evidence of intended delegation. Because current agentic systems lack independent goals, their behavior is determined by what the user specifies [Xing et al., 2026]: a goal, a procedure for reaching it, and the tools or information needed for repeated execution [Ferrag et al., 2026, Gupta and Kumar, 2026]. A configuration therefore records not merely that a practitioner used AI, but that they built it into how a task gets done. Delegation is therefore directly observable in agentic settings, whereas in conversational use it must be inferred from exchanges that rarely reveal whether a task was handed off or merely assisted. Agentic systems also reach further into work than prompt-level LLM use. Rather than producing a single output, an agent shifts the practitioner from performing each step to supervising a whole sequence of work [Yang et al., 2026]. Despite this change, prior work has mainly focused on the capabilities of agents [Shao et al., 2026, Gupta and Kumar, 2026] or their availability [Sarkar, 2026], rather than on delegated exposure. Studies that observe real usage logs, meanwhile, do not link them to the O*NET task structure [Johnston et al., 2026], which makes their findings hard to compare with prior exposure measures. We therefore measure delegated exposure with the Agentic Adoption Index (AAI), built from practitioner-written descriptions of configured agents, linked to 4 Figure 1: Four nested layers of AI exposure: capability (what AI can do), availability (what is made usable), observed (what workers use), and delegated (what workers commit to AI). Each layer is nested within the one before it. O*NET task statements, and aggregated to the occupation level. Because a configuration reflects a commitment already made rather than a possibility, delegated exposure should track where AI is most likely to change how work is done. 3 Data 3.1 Practitioner-configured agent skills Measuring delegated exposure calls for a different kind of data than the prompt-level conversation logs used in prior observed-exposure work. It requires evidence of how users configure AI agents to perform work, and such evidence should satisfy two conditions. First, users themselves should author it, so that it reflects their own configuration choices directly. Second, it should specify what an agent is intended to do in task-like terms, preferably at a level of abstraction comparable to occupational task descriptions as in O*NET. The skill files published on the Manus Skills Marketplace 1 satisfy both conditions. The Mar- 1 Manus AI, developed by the Singapore-based startup Butterfly Effect Technology, is an autonomous AI agent that executes multi-step tasks on a virtual computer. Their Skills Marketplace aggregates skill.md files published to public open-source repositories, primarily GitHub, where practitioners share them much as they share any open- source code—to distribute reusable automation recipes and to demonstrate what agentic workflows can accomplish. As of mid-2026, it indexes over 2.6 million skills drawn from more than 320,000 repositories contributed by roughly 230,000 creators. These contributions come predominantly from technically sophisticated early adopters, yet span nearly the full range of occupations, from highly specialized tasks to routine ones. This breadth of openly shared, contributor-driven skills, rather than the platform’s raw usage volume, is what makes the Marketplace a useful window onto which occupational tasks practitioners are actually automating. 5 ketplace is a public repository where practitioners themselves author and publish self-contained skill.md files. A skill is a modular resource that encapsulates a specific function or workflow, providing the agent with detailed instructions for performing specialized tasks. 2 Each skill.md file consists of a short description stating what the skill does and when it should be invoked, together with a body containing the full execution instructions. Because the description states what is done rather than how, it sits at the same level of abstraction as an O*NET task statement, which makes it a natural unit for semantic comparison with occupational tasks. We collect accessible skill.md files from the marketplace and, after removing duplicates, obtain 117,887 distinct skills. We preprocess these in two steps. First, we remove skills that contain a description but no execution instructions. We also exclude skills whose description is under 100 characters, as such descriptions are too short to convey the automation context needed for a meaningful semantic match. This leaves 56,532 skills. Second, we filter the remaining skills for work-relatedness using GPT-4o-mini. We prompt the model to classify each description to classify each description as work-related (a professional skill, workflow, tool usage, or occupational task) or not (non-English, or unrelated to any work context, e.g. a local travel guide or a home recipe manager), retaining only the former (See SI Section 1 for the full prompt and settings). This retains 53,515 as our analysis sample. 3.2 Occupational tasks and characteristics To connect agent skills to the occupations whose work they could automate, we need data on the tasks that make up each occupation’s work. We draw on O*NET [National Center for O*NET Development, 2025], a database developed and maintained by the US Department of Labor, using O*NET 30.2, released in 2025. It provides standardized, occupation-specific descriptions of the knowledge, skills, and tasks required across occupations, and remains the most widely used source for occupation-level task analysis. Among these, we focus on the task descriptors because agent skills are themselves specifications of discrete tasks to be executed, making tasks the level at which the two can be directly compared. For each task, O*NET’s Task Ratings report an importance score on a five-point scale, averaged across surveyed respondents. When aggregating task-level exposure scores to the occupation level, we use this mean importance as the weight for each task, so that tasks more central to an occupation contribute more to its occupation-level score. After excluding tasks without a valid importance rating, we retain 17,951 of the 18,797 O*NET task statements. This exclusion also removes 26 occupations whose tasks lack ratings entirely. When we compute AAI scores at the task level and aggregate them to the occupation level, we keep only the base O*NET-SOC codes (those ending in .00), dropping their finer subdivisions (e.g., 11-1011.03), so that the occupation units align with the existing exposure measures and the wage and education data used in our analysis. This yields 748 occupations for analysis. Finally, to examine how agentic adoption varies with occupational earnings and education, we draw on data on occupational wages, employment, and education from the Bureau of Labor Statistics Occupational Employment and Wage Statistics [Bureau of Labor Statistics, 2025]. For each occupation, we use the 2025 annual median wage, national employment level, and typical entry-level education requirement defined as the education level most commonly required to enter the occupation as designated by occupational experts. This information is available for 560 of the 748. 2 Throughout this paper, we use typewriter font to denote this technical, platform-specific sense of the term, distinguishing it from the broader notion of “skill” used in the human capital literature. 6 3.3 Existing exposure measures To understand how delegated exposure differs from previous measures, we compare it against exist- ing measures from capability, availability, and observed exposure. We begin with a pre-AI baseline, the probability of computerisation [Frey and Osborne, 2017], which estimates occupations’ automa- tion risk before large language models emerged and lets us test whether agentic diffusion follows the same routine-task patterns that earlier automation frameworks anticipated. We then turn to recent AI-exposure research and select measures according to three criteria. Each measure should (1) target current AI systems such as LLMs and agents, (2) adopt a distinct methodological approach spanning the capability, availability, and observed layers introduced in Section 2.1, and (3) report scores comparable at the occupation level. Guided by these criteria, we select three complementary measures, one per layer. The RL Feasibility Index of Tomei and Teeselink [2026] captures capability exposure by scoring occupational tasks on their suitability for reinforcement-learning-based automa- tion with Gemini 2.5 Flash. The task-level exposure ratings of Eloundou et al. [2024], (γ), capture availability exposure by assessing whether an LLM combined with complementary software could reduce task completion time by at least 50%. The usage-based measure of Massenkoff and McCrory [2026] captures observed exposure based on actual Claude usage. All four measures, including the pre-AI baseline, are obtained directly from data released by their respective authors. 4 Methods 4.1 Measuring task–skill similarity Measuring how closely agent skills correspond to occupational tasks requires a way to quantify semantic similarity between two large, unstructured collections of short text: the skill descriptions and the O*NET task statements. Embedding text into a shared vector space and measuring cosine similarity between texts is a common approach for this kind of matching in social science [Matsui and Ferrara, 2024]. Following this approach, we embed every skill description and O*NET task state- ment into a common vector space using all-MiniLM-L6-v2, a Sentence-BERT model optimized for semantic similarity tasks that maps text to 384-dimensional dense vectors [Reimers and Gurevych, 2019]. Although the Manus Skills Marketplace classifies approximately 90% of skills by O*NET occupation code, we do not use these pre-assigned labels in our analysis. They tie each skill to a predefined occupation, whereas a skill often supports tasks spanning several occupations, and our goal is to measure this cross-occupational correspondence rather than rely on the marketplace’s occupation labels. Instead, we compute each skill’s semantic similarity against every O*NET task statement, so that a single skill can match tasks across multiple occupations (see Figure 2 for a schematic summary). We compute the cosine similarity between each task t and skill s as sim(t,s) = V t ·V s ∥V t ∥V s ∥ ,(1) whereV t andV s are unit-normalized embedding vectors (See Tables S1 and S2 in SI Section 2 for the highest- and lowest-similarity tasks). 4.2 Constructing the AAI Having measured task–skill similarity, we aggregate these pairwise scores into a single index for each occupation. For each task t, we compute a task-level exposure score by averaging its cosine similarity across all skill descriptions in the corpus. We then aggregate to the occupation level using importance-weighted averaging over all tasks associated with occupation j: 7 Figure 2: Example structure of a skill.md file and its mapping to O*NET tasks. Each skill consists of a name, a short description, and a body containing full execution instructions. The description is embedded and compared against all O*NET task statements via cosine similarity, yielding a task-skill similarity matrix from which the occupation-level AAI is aggregated. Agentic Adoption Index j = X t∈T j w jt · 1 |S| X s∈S sim(t,s),(2) where T j is the set of O*NET task statements associated with occupation j, S is the set of all retained skill descriptions with |S| = 53,515, and w jt is the importance weight of task t within occupation j, normalized to sum to one so that occupations with more tasks are not scored higher by construction. A higher AAI indicates that an occupation’s tasks are, on average, well covered by existing agentic skills that early-adopting practitioners have already built and published. As a robustness check, we replicate the index using a different sentence embedding model, all-mpnet-base-v2, in place of all-MiniLM-L6-v2. The two occupation-level indices exhibit a Pearson correlation of 0.91, suggesting that the results are not sensitive to the choice of sentence embedding model. 4.3 Validating the AAI We explore score highest and lowest on the AAI, which shows where agentic skills align most and least closely with an occupation’s tasks. Tables 1 and 2 report the top and bottom 20 occupations by AAI. High-AAI occupations are information-intensive roles built around analysis, documenta- tion, and coordination, such as management analysts, technical writers, and computer programmers. Low-AAI occupations require manual dexterity or direct physical intervention, such as oral and maxillofacial surgeons, roofers, and tire builders. To characterize this pattern, we relate the AAI to a cognitive ability share constructed from O*NET ability importance scores. The AAI is positively correlated with this share (See Figure S1 in SI Section 3), confirming that high-AAI occupations rely on cognitive rather than physical or psychomotor abilities. 8 Table 1: Top 20 occupations by the Agentic Adoption Index OccupationAAI Management Analysts0.192 Technical Writers0.182 Natural Sciences Managers0.176 Computer Programmers0.175 Production, Planning, and Expediting Clerks0.172 Survey Researchers0.170 Statistical Assistants0.170 Industrial Engineers0.169 First-Line Supervisors of Office and Administrative Support Workers 0.168 Social Science Research Assistants0.168 Computer and Information Systems Managers0.167 Administrative Services Managers0.167 Architectural and Engineering Managers0.167 Electronics Engineers, Except Computer0.166 Computer User Support Specialists0.165 Software Developers0.164 Office Clerks, General0.163 Training and Development Specialists0.162 First-Line Supervisors of Construction Trades and Extraction Workers 0.161 Industrial Engineering Technologists and Technicians0.160 Table 2: Bottom 20 occupations by the Agentic Adoption Index OccupationAAI Oral and Maxillofacial Surgeons0.041 Tapers0.041 Automotive Glass Installers and Repairers0.047 Podiatrists0.047 Roofers0.048 Tire Builders0.054 Prosthodontists0.056 Shampooers0.057 Dental Hygienists0.058 Floor Sanders and Finishers0.060 Dermatologists0.061 Tire Repairers and Changers0.061 Embalmers0.062 Slaughterers and Meat Packers0.062 Roof Bolters, Mining0.063 Cement Masons and Concrete Finishers0.063 Bicycle Repairers0.064 Animal Breeders0.064 Tile and Stone Setters0.066 Carpet Installers0.066 9 4.4 Analytic strategy We examine the properties of AAI through three complementary analyses, each addressing a different question about what AAI captures. First, we ask whether the occupations that earlier, pre-AI frameworks rated as high in automation risk are the ones that now show high agentic adoption. We address this by comparing AAI with the pre-AI computerisation risk of Frey and Osborne [2017]. Second, we ask how agentic adoption relates to existing AI-exposure measures. These measures mostly capture general LLM use in services such as ChatGPT and Claude, but not agentic use. We address this by comparing AAI with three exposure measures spanning the capability, availability, and observed layers introduced in Section 2.1. Third, we ask how occupations at different wage and education levels adopt agentic AI, using employment-weighted least squares regressions. We explain each analytic approach in turn. 4.4.1 The AAI and pre-AI automation risk We examine whether the occupations that earlier, pre-AI frameworks rated as high in automation risk are the ones now adopting agentic AI most heavily. To answer this, we compare AAI with the probability of computerisation of Frey and Osborne [2017], a widely cited occupation-level measure of automation risk. Because that measure classifies occupations as high- or low-risk, we split occupations at the median of AAI into high- and low-AAI groups. We then measure the overlap between the high-risk and high-AAI groups using the Jaccard similarity coefficient. A low Jaccard value would indicate that the occupations most exposed under pre-AI automation frameworks are largely not the ones adopting agentic AI most heavily today. 4.4.2 The AAI and existing exposure measures Having introduced the AAI as a configuration-level extension of observed exposure, we compare it empirically with the capability, availability, and observed layers described in Section 2.1. We use the three exposure measures introduced in Section 3.3, one for each layer, to see which AAI tracks most closely. Table 3 reports summary statistics for these three measures and the AAI. The AAI has a mean and median of about 0.11 and a narrower range (0.04 to 0.19) than the others. The availability exposure γ of Eloundou et al. [2024] spans the full 0 to 1 range with a median of 0.43, the Claude usage measure of Massenkoff and McCrory [2026] concentrates near zero (median 0.01) with a long right tail up to 0.51, and the RL Feasibility Index of Tomei and Teeselink [2026] ranges from 0 to about 69 with a median of 23.60. These differences in scale motivate the rank-based comparisons used in the analyses that follow. 4.4.3 The AAI, wages, and education We examine how occupations at different wage and education levels adopt agentic AI through a series of employment-weighted least squares regressions with the occupation-level AAI as the dependent variable. We weight each occupation by its 2025 employment, so that the results describe patterns across the workforce rather than across job titles. Model (1) asks whether the AAI keeps rising with wages and education, reaching its highest values in the best-paid and most-educated occupations, or instead peaks in the middle of the wage and education range. To allow for both possibilities, we regress the AAI of occupation j on the logarithm of the 2025 median annual wage w j and its square, together with education dummies for high school or below (D HS j ) and master’s or above (D MA j ), relative to a bachelor’s degree, 10 Table 3: Descriptive statistics for the Agentic Adoption Index and existing exposure measures. Agentic Adoption Index Tomei and Teeselink (2026) (Gemini) Eloundou et al. (2024) (GPT-4) Massenkoff and McCrory (2026) (Claude) N † 748606618511 Mean0.1124.830.480.02 Min0.040.000.000.00 25th pct0.1012.240.140.00 Median0.1123.600.430.01 75th pct0.1335.960.850.02 Max0.1969.421.000.51 † Occupation counts vary slightly across measures because each source relies on a different O*NET release and level of measurement (e.g., tasks, detailed work activities, or skill-level descriptions), which map to occupations with slightly different coverage. Model (1): AAI j = β 0 + β 1 logw j + β 2 (logw j ) 2 + β 3 D HS j + β 4 D MA j + ε j .(3) Model (2) turns to the availability of AI, asking whether occupations whose tasks current AI tools can already handle are also those for which practitioners have built the most agent skills. We regress the AAI on the availability exposure γ j of Eloundou et al. [2024] alone, Model (2): AAI j = β 0 + β 1 γ j + ε j .(4) Model (3) brings the two together, examining whether the association between the AAI and γ j remains significant once wages and education are added as controls, and, conversely, whether wages and education remain significant once γ j is held constant, Model (3): AAI j = β 0 + β 1 γ j + β 2 logw j + β 3 (logw j ) 2 + β 4 D HS j + β 5 D MA j + ε j .(5) With these three models, we compare the coefficients across them and interpret the differences. 5 Results 5.1 Agentic adoption does not follow pre-AI automation risk We compare the AAI with the pre-AI automation risk of Frey and Osborne [2017] to assess whether the occupations now adopting agentic AI are those that earlier frameworks expected to be auto- mated. Plotted against the probability of computerisation, the AAI shows little correspondence with it (Figure 3). Occupations that earlier frameworks grouped together by automation risk span a wide range of AAI values. Among occupations with low pre-AI automation risk, some score low on AAI, such as oral surgeons, prosthodontists, and podiatrists, while others score high, such as management analysts, industrial engineers, and natural sciences managers. Occupations with high pre-AI automation risk divide in the same way, with manual roles such as roofers and tire builders scoring low and information-intensive ones such as technical writers and statistical assistants scoring high. To quantify this divergence, we split occupations at the median of AAI into high and low groups, mirroring the high- and low-risk partition of the Frey–Osborne measure, and assess the agreement between the two classifications with the Jaccard similarity coefficient. The coefficient is only 0.05, indicating that the occupations flagged as high-risk by the pre-AI measure are for the most part distinct from those now exhibiting the highest AAI. This weak overlap between pre-AI risk and AAI also holds at the sector and state levels (See Figures S2-S4 in SI Section 4). 11 Figure 3: AAI versus Computerization Risk across occupations. The x-axis is AAI; the y-axis is the probability of computerization from Frey and Osborne [2017]. Occupations with sim- ilar computerization risk diverge widely in the AAI, so the two measures capture different sets of occupations. 5.2 The AAI aligns with capability more than with observed use Because the AAI introduces a configuration-level view of exposure, we situate it relative to estab- lished measures, each of which captures a distinct layer of how much AI can affect an occupation. We compare the AAI against the three measures of Section 3.3, one per layer, to identify the layer with which agentic adoption corresponds most closely to. All three Spearman correlations are pos- itive and significant (Figure 4), indicating that our practitioner-derived index is broadly consistent with prior work. The correlations differ in strength, however. The AAI correlates about equally with the capability measure of Tomei and Teeselink [2026] (ρ = 0.673 ∗ ) and the availability mea- sure of Eloundou et al. [2024] (ρ = 0.623 ∗ ), with only a small gap between the two, whereas its correlation with the observed usage measure of Massenkoff and McCrory [2026] (ρ = 0.374 ∗ ) is comparatively weaker. The AAI thus aligns more closely with AI capability and availability than where it is already deployed in practice. This ordering across the three measures suggests that, early in the diffusion of agentic AI, the AAI captures the potential for agents to take over an occupation’s tasks rather than the extent to which that occupation currently uses AI. 5.3 Availability explains adoption, except at the top To see how agentic adoption is positioned across the labor market, we examine the regression results from Section 4.4.3. Model (1) shows that the AAI peaks below the top of the wage distribution and declines at both extremes. Across wages, AAI follows an inverted-U pattern, with a positive linear term (β = 0.447 ∗ ) and a negative quadratic term (β =−0.019 ∗ ). When it comes to education, both “high school or below” (β =−0.011 ∗ ) and “master’s or above” (β =−0.026 ∗ ) occupations 12 Figure 4: Correlation between AAI and established exposure measures. Each panel plots the AAI against one of three occupation-level exposure scores: the Gemini-based capability exposure (RL Feasibility Index) from [Tomei and Teeselink, 2026] (left), ChatGPT-based availability exposure γ from [Eloundou et al., 2024] (center), and Claude-based observed exposure from [Massenkoff and McCrory, 2026] (right). The regression line is fitted in linear space for the left panel and in log-x space for the center and right panels. have significantly lower AAI than bachelor’s-degree occupations (Table 4), so the AAI is highest at the bachelor’s level. Model (2) relates the AAI to the availability of AI to see whether users adopt AI agents when the service becomes commercialized. On its own, the availability exposure γ enters positively and significantly (β = 0.055 ∗ ) and alone explains 57.8% of the variance in AAI (R 2 = 0.578), far more than wages and education together in Model (1) (R 2 = 0.349). Agentic adoption thus depends mainly on whether current AI tools can already handle an occupation’s tasks, rather than on wages or education. Model (3) includes wages, education, and γ together. The availability exposure γ remains significant (β = 0.046 ∗ ), so its association with AAI suggests the tendency of high-availability occupations to be better paid and more educated. The wage terms also remain significant (β = 0.201 ∗ for log wage and β = −0.009 ∗ for its square), though both are much smaller than in Model (1) (0.447 ∗ and −0.019 ∗ ). The inverted-U shape therefore holds, but becomes weaker once γ is included, so part of the wage effect in Model (1) reflects the availability of AI. For occupations requiring high school or below, the gap from bachelor’s-degree occupations narrows from −0.010 ∗ in Model (1) to −0.002 and loses significance, indicating that their lower AAI reflects tasks that current AI tools cannot yet handle rather than their education itself. For occupations requiring a master’s degree or above, the gap remains significant and changes little, from −0.026 ∗ to −0.018 ∗ , so availability alone does not account for their lower AAI. The lower AAI of less-educated occupations thus follows the reach of current AI tools, whereas that of the most-educated occupations does not. Adding wages and education raises R 2 only modestly, from 0.601 to 0.621, so socioeconomic position contributes little to explaining agentic adoption once availability is accounted for. What makes the master’s-degree gap notable is not its contribution to fit but its persistence: it survives the inclusion of γ while the corresponding gap for less-educated occupations does not. This points to a barrier to agentic adoption in the most-educated occupations that lies beyond availability. Figure 5 summarizes this pattern. The left panel of the figure shows that, with availability 13 exposure and education included in the model, the AAI follows an inverted-U shape in wages. The right panel reports the estimated education effects from the same model. Relative to a bachelor’s degree, occupations requiring a master’s or above have significantly lower AAI, while the gap for high school or below is statistically not significant. Taken together, the figure confirms that availability accounts for part of the inverted-U wage relationship, while for education it does not explain the persistent gap among the most-educated occupations. Table 4: WLS regressions of Agentic Adoption Index. Standard errors in parentheses. (1)(2)(3) Intercept−2.497 ∗ 0.092 ∗ −1.081 ∗ (0.521)(0.001)(0.404) Availability Exposure (γ)0.055 ∗ 0.046 ∗ (0.002)(0.002) log wage0.447 ∗ 0.201 ∗ (0.094)(0.073) (log wage) 2 −0.019 ∗ −0.009 ∗ (0.004)(0.003) High school or below (ref. Bachelor’s)−0.011 ∗ −0.002 (0.003)(0.002) Master’s or above (ref. Bachelor’s)−0.026 ∗ −0.018 ∗ (0.005)(0.004) R 2 0.3490.5780.621 Adj. R 2 0.3440.5780.618 N560593560 ∗ p < .1, ∗ p < .05, ∗ p < .01 Alternative exposure measures show a similar pattern (See Table S3 in SI Section 5). The RL Feasibility Index of Tomei and Teeselink Tomei and Teeselink [2026], a capability-type measure, explains 60% of the variance in AAI once wages and education are added (R 2 = 0.605), about the same as the availability exposure used in the main models. The Claude usage measure of Massenkoff and McCrory [2026], an observed-type measure, explains far less even with education included (R 2 = 0.249). The AAI therefore reflects what early adopters have already built as agentic workflows, rather than how widely AI is used across the broader workforce. The wage terms remain significant across every specification that includes them, and occupations requiring a master’s degree or above stay below bachelor’s-degree occupations throughout, by −0.017 ∗ to −0.026 ∗ . Taken together, the three models point to one conclusion. People delegate their work to agents mostly when current AI tools can already handle their work, and less by wages or education levels. The AAI peaks in the upper-middle of the wage and education range, but this peak follows the availability of AI to those occupations rather than their wage or education level in itself. Once availability is accounted for, the lower AAI of less-educated occupations loses significance, since these are the occupations whose work current AI tools cannot yet handle. What remains is at the other end. Occupations requiring a master’s degree or above adopt agentic AI less than occupations at similar levels of availability, a gap that availability does not explain. 14 Figure 5: The AAI’s relationships with wage and education. Both panels are based on the full WLS regression in Model (3) of Table 4. (Left) Predicted AAI across annual wage, holding education at the Bachelor’s degree reference group and γ at its employment-weighted mean. Each point is an occupation, with dot size proportional to employment, and the solid line is the model-predicted AAI with the shaded band denoting its 95% confidence interval. The marked point indicates the wage at which predicted AAI peaks. (Right) Predicted AAI for each education level, holding log wage and γ at their employment-weighted means. Points are the model-predicted values and horizontal whiskers denote 95% confidence intervals. 6 Discussion and Conclusion We extend research on AI exposure by adding the concept of delegated exposure, as LLM use moves beyond isolated conversations toward reusable agentic workflows. We operationalize it through the AAI, which matches skill descriptions from the Manus Marketplace to O*NET task statements. The resulting measure reveals a distinct occupational gradient. Delegated exposure concentrates in information-intensive work and remains minimal where work depends on manual dexterity or direct intervention in the physical environment. This distribution departs from pre-AI accounts such as computerization-risk estimates, which located automation risk in routine tasks and ranked occupa- tions differently from what we observe. The AAI also aligns more closely with measures of technical capability and application availability than with prompt-level conversational use, suggesting that it tracks what AI can in principle perform rather than what usage records currently capture. Yet the availability measure does not explain the pattern fully: adoption falls below what availability expo- sure would predict at both ends of the wage distribution, and among highly educated occupations independent of wage. Our findings show that occupational AI exposure needs to be interpreted in relation to the stage and form of diffusion under examination. Measures of capability, availability, and observed exposure are not interchangeable readings of a single underlying quantity; each captures a distinct phase, from what a technology can do, to what has been built with it, to what people take up. Disagreements in this literature may therefore partly reflect measures of different layers being compared as though they were the same thing. The AAI adds a further layer by identifying where practitioners have begun configuring AI systems to automate their own work, and its divergence from observed exposure is telling. A prompt records a single request, whereas an agent description specifies the role, procedure, and 15 sequence of activities a practitioner intends to automate. Because that specification takes effort, practitioners undertake it only where the technology can dependably do the work, which is why delegated exposure resembles capability exposure more than observed exposure does. Conversational records, in turn, may understate how far practitioners have committed to automating work. Our models show that technical availability accounts for most of the variation in agent adop- tion. This complicates a well-established expectation that use lags capability: general purpose technologies typically deliver returns only after firms make the complementary investments in pro- cess redesign, training, and reorganization that let the technology be absorbed [Taylor and Helfat, 2009, Brynjolfsson and Milgrom, 2013, Machkour and Abriane, 2026], and technologies are reshaped by the settings that take them up [Klein and Kleinman, 2002, Berker et al., 2005]. Our results sug- gest those adjustments may play a smaller role than expected, at least in the early adoption of agentic systems. Two features of agentic systems may explain this pattern. First, these systems are configured in natural language, so for the practitioners in our data the distance between recognizing that a task could be automated and building something that does it is short. The skill files we observe were written and published by individuals, without the procurement, integration, or specialized expertise that would ordinarily mediate a workplace technology [Bright et al., 2025]. Second, generality means the technology adapts to the task rather than the task to the technology, removing much of the mutual adjustment that typically slows adoption. Yet this pattern is not uniform. The AAI follows an inverted-U across wages, concentrating neither in low-wage work nor in the highest-paid professions, and across education it peaks at the bachelor’s level and declines among occupations requiring a graduate degree. Availability accounts for the low end: less-educated occupations adopt less because current tools cannot yet handle their work. It does not account for the top. One explanation is that work at the top depends on tacit expertise, contextual judgment, and intensive interpersonal interaction, none of which is easily specified in advance [Autor, 2015]. But our data cannot distinguish constraint from choice. These occupations may be unable to specify their work in advance, or they may be able to and decline, since professionals have both the autonomy and the incentive to control the pace at which their work is codified [Abbott, 1988]. The AAI therefore captures not where automation has already occurred, but where early practitioners are actively attempting to automate their own work. Longitudinal evidence would help here, and not only here. If the shortfall at the top narrows as agentic systems improve, constraint is the better explanation; if it persists while capability rises, the case for choice strengthens. Pre-AI computerisation estimates were influential within a decade of publication and are already dated, which suggests that any single-period measure of exposure has a short shelf life. Tracking each layer over time would show not only how exposure grows but how the layers move in relation to one another. Sustaining such tracking over time faces a structural obstacle. Conversational usage data sits with a small number of platform firms, and expert annotation of tasks is costly and slow to up- date. Publicly shared artifacts avoid both constraints. The skill files we analyze were published by practitioners for other practitioners, which makes them observable without platform cooperation and renewable as the marketplace grows. Their availability is not guaranteed, since it depends on norms of open sharing that platforms may not sustain as the technology commercializes. But our data source suggests that the traces left by users configuring their own tools may provide a more accessible basis for tracking diffusion over time. Several limitations qualify these conclusions, suggesting directions for future research. First, the agent corpus likely overrepresents technically sophisticated, highly engaged practitioners, so the AAI should be read as an indicator of early-adopter activity rather than of adoption across the workforce. Tracking it longitudinally as platforms mature would show whether the present concen- 16 tration persists or spreads to other areas of work. Second, the AAI captures whether practitioners have configured an agent for a task, not how the agent structures and carries out the work. An agent may decompose or sequence a task in ways that differ substantially from how a human performs it. Examining execution traces or the internal structure of agent runs, rather than skill descriptions alone, would clarify how agentic task performance diverges from human work. Third, the analysis relies on a static list of O*NET tasks. As Acemoglu and Restrepo [2019] note, the full effect of automation involves both a displacement effect, in which machines substitute for existing tasks, and a reinstatement effect, in which new tasks emerge. The AAI speaks only to the former and cannot represent the new work that agents create. Capturing reinstatement will require task inventories that update as new work emerges, drawn from sources such as job postings or agent operation records. Such an approach would reveal not only where work is being automated but also where new forms of human work are being created. 17 References Andrew Abbott. The system of professions: An essay on the division of expert labor. University of Chicago press, 1988. Daron Acemoglu and Pascual Restrepo. Automation and new tasks: How technology displaces and reinstates labor. Journal of economic perspectives, 33(2):3–30, 2019. David H Autor. Why are there still so many jobs? the history and future of workplace automation. Journal of economic perspectives, 29(3):3–30, 2015. Thomas Berker, Maren Hartmann, Yves Punie, and Katie Ward. Domestication of media and technology. McGraw-Hill Education (UK), 2005. Alexander Bick, Adam Blandin, and David J Deming. The rapid adoption of generative AI. Man- agement Science, 2026. Jonathan Bright, Florence Enock, Saba Esnaashari, John Francis, Youmna Hashem, and Deborah Morgan. Generative ai is already widespread in the public sector: evidence from a survey of uk public sector professionals. Digit. Gov.: Res. Pract., 6(1), February 2025. doi: 10.1145/3700140. URL https://doi.org/10.1145/3700140. Erik Brynjolfsson and Paul Milgrom. Complementarity in organizations. The handbook of organi- zational economics, pages 11–55, 2013. Erik Brynjolfsson, Tom Mitchell, and Daniel Rock. What can machines learn and what does it mean for occupations and the economy? AEA Papers and Proceedings, 108:43–47, 2018. doi: 10.1257/pandp.20181019. Bureau of Labor Statistics. Occupational employment and wage statistics (OEWS), may 2025, 2025. URL https://w.bls.gov/oes/. Ayush Chopra, Santanu Bhattacharya, DeAndrea Salvador, Ayan Paul, Teddy Wright, Aditi Garg, Feroz Ahmad, Alice C. Schwarze, Ramesh Raskar, and Prasanna Balaprakash. The iceberg index: Measuring skills-centered exposure in the AI economy, 2025. URL https://arxiv.org/ abs/2510.25137. Gustavo de Souza. Ai in the office and the factory: Evidence from administrative software registry data. Working Paper 2025-11, Federal Reserve Bank of Chicago, June 2026. URL https://ssrn. com/abstract=5375463. R. Maria del Rio-Chanona, Ekkehard Ernst, Rossana Merola, Daniel Samaan, and Ole Teutloff. AI and jobs. a review of theory, estimates, and evidence, 2025. URL https://arxiv.org/abs/ 2509.15265. G. Demirev. Ai product innovation and occupational exposure: Automation and augmentation in commercial ai deployments. Industry and Innovation, pages 1–30, 2026. doi: 10.1080/13662716. 2026.2623903. Tyna Eloundou, Sam Manning, Pamela Mishkin, and Daniel Rock. GPTs are GPTs: Labor market impact potential of LLMs. Science, 384(6702):1306–1308, 2024. doi: 10.1126/science.adj0998. 18 Edward Felten, Manav Raj, and Robert Channing Seamans. The effect of artificial intelligence on human labor: An ability-based approach. Academy of Management Proceedings, 2019(1):15784, 2019. doi: 10.5465/AMBPP.2019.140. Edward Felten, Manav Raj, and Robert Seamans. Occupational, industry, and geographic exposure to artificial intelligence: A novel dataset and its potential uses. Strategic Management Journal, 42(12):2195–2217, 2021. doi: 10.1002/smj.3286. Edward W. Felten, Manav Raj, and Robert Seamans. A method to link advances in artificial intelligence to occupational abilities. AEA Papers and Proceedings, 108:54–57, 2018. doi: 10. 1257/pandp.20181021. Edward W. Felten, Manav Raj, and Robert Seamans. How will language modelers like ChatGPT affect occupations and industries? SSRN Electronic Journal, 2023. doi: 10.2139/ssrn.4375268. Enrico Maria Fenoaltea, Dario Mazzilli, Aurelio Patelli, Angelica Sbardella, Andrea Tacchella, An- drea Zaccaria, Marco Trombetti, and Luciano Pietronero. Follow the money: A startup-based measure of ai exposure across occupations, industries, and regions. PNAS Nexus, 5(6), May 2026. ISSN 2752-6542. doi: 10.1093/pnasnexus/pgag185. URL http://dx.doi.org/10.1093/ pnasnexus/pgag185. Mohamed Amine Ferrag, Norbert Tihanyi, and Mérouane Debbah. From LLM reasoning to autonomous AI agents: A comprehensive review. IEEE Access, 14:84237–84285, 2026. doi: 10.1109/ACCESS.2026.3698694. Morgan R. Frank, David Autor, James E. Bessen, Erik Brynjolfsson, Manuel Cebrian, David J. Deming, Maryann Feldman, Matthew Groh, José Lobo, Esteban Moro, Dashun Wang, Hyejin Youn, and Iyad Rahwan. Toward understanding the impact of artificial intelligence on labor. Proceedings of the National Academy of Sciences, 116(14):6531–6539, 2019. doi: 10.1073/pnas. 1900949116. URL https://w.pnas.org/doi/abs/10.1073/pnas.1900949116. Morgan R. Frank, Alireza Javadian Sabet, Lisa Simon, Sarah H. Bana, and Renzhe Yu. AI-exposed jobs deteriorated before ChatGPT, 2026. URL https://arxiv.org/abs/2601.02554. Carl Benedikt Frey and Michael A Osborne. The future of employment: How susceptible are jobs to computerisation? Technological forecasting and social change, 114:254–280, 2017. Ravish Gupta and Saket Kumar. Agentic AI and occupational displacement: A multi-regional task exposure analysis of emerging labor market disruption, 2026. URL https://arxiv.org/abs/ 2604.00186. Kunal Handa, Alex Tamkin, Miles McCain, Saffron Huang, Esin Durmus, Sarah Heck, Jared Mueller, Jerry Hong, Stuart Ritchie, Tim Belonax, et al. Which economic tasks are performed with AI? evidence from millions of claude conversations. arXiv preprint arXiv:2503.04761, 2025. Drew Johnston, David Holtz, Alex Martin Richmond, Christopher Ong, Prasanna Tambe, and Aaron Chatterji. The shift to agentic AI: Evidence from codex, 2026. URL https://arxiv.org/ abs/2606.26959. Hans K Klein and Daniel Lee Kleinman. The social construction of technology: Structural consid- erations. Science, Technology, & Human Values, 27(1):28–52, 2002. 19 Badr Machkour and Ahmed Abriane. Artificial intelligence in organizations: A systematic review of operationalizing generative artificial intelligence capabilities for organizational performance. IEEE Access, 2026. Maxim Massenkoff and Peter McCrory. Labor market impacts of AI: A new measure and early evidence, 2026. URL https://w.anthropic.com/research/labor-market-impacts. Akira Matsui and Emilio Ferrara. Word embedding for social sciences: An interdisciplinary survey. PeerJ Computer Science, 10:e2562, 2024. Luca Mouchel, Pierre Bouquet, and Yossi Sheffi. Jobs’ AI exposure should be measured from evidence, not model priors, 2026. URL https://arxiv.org/abs/2605.15474. National Center for O*NET Development. O*NET 30.2 database, 2025. URL https://w. onetcenter.org. Nils Reimers and Iryna Gurevych. Sentence-BERT: Sentence embeddings using siamese BERT- networks. In Proceedings of the 2019 conference on empirical methods in natural language process- ing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP), pages 3982–3992, 2019. Soumyajit Sarkar. Enterprise usage of AI agents: A comprehensive survey on applications, frame- works, and future productivity impact. Frameworks, and Future Productivity Impact (March 09, 2026), 2026. Yijia Shao, Humishka Zope, Yucheng Jiang, Jiaxin Pei, David Nguyen, Erik Brynjolfsson, and Diyi Yang. Future of work with AI agents: Auditing automation and augmentation potential across the u.s. workforce, 2026. URL https://arxiv.org/abs/2506.06576. Alva Taylor and Constance E Helfat. Organizational linkages for surviving technological change: Complementary assets, middle management, and ambidexterity. Organization Science, 20(4): 718–739, 2009. Philip Moreira Tomei and Bouke Klein Teeselink. What jobs can AI learn? measuring exposure by reinforcement learning. arXiv preprint arXiv:2605.02598, 2026. Kiran Tomlinson, Sonia Jaffe, Will Wang, Scott Counts, and Siddharth Suri. Working with AI: Measuring the applicability of generative AI to occupations, 2025. Eric Xing, Mingkai Deng, and Jinyu Hou. Critique of agent model, 2026. URL https://arxiv. org/abs/2606.23991. Jeremy Yang, Noah Yonack, Kate Zyskowski, Denis Yarats, Johnny Ho, and Jerry Ma. The adoption and usage of AI agents: Early evidence from perplexity. Working Paper 26-040, Harvard Business School, December 2025. URL https://ssrn.com/abstract=5887343. Jeremy Yang, Kate Zyskowski, Noah Yonack, and Jerry Ma. How AI agents reshape knowledge work: Autonomy, efficiency, and scope. arXiv preprint arXiv:2606.07489, 2026. Michelle Yin and Burhan Ogut. Who uses AI? platform selection and the measurement of occupa- tional AI exposure, 2026. URL https://arxiv.org/abs/2605.21743. 20