1. Working with Agentic AI: Data Scientist

How human skills and agentic skills are dynamically integrated and coordinated across data science responsibilities

1.1. Data Scientist as an Occupation

A data scientist collects, prepares, analyzes, models, interprets, and communicates data to answer complex questions and support evidence-based decisions. This occupation connects statistical reasoning, machine learning, data engineering, domain understanding, and communication by transforming raw data into analytical findings, predictive models, experimental evidence, and decision-support outputs. To ensure that this work produces reliable value, the data scientist must understand how problem definitions, data sources, modeling methods, evaluation criteria, technical constraints, stakeholder needs, and organizational consequences collectively determine the meaning and usefulness of analytical results.

Typical responsibilities of a data scientist include:

• analytical problem framing and data science requirement definition;
• data acquisition, preparation, and feature engineering;
• exploratory data analysis and statistical pattern identification;
• machine learning model development and optimization;
• model validation, testing, and performance evaluation;
• experimentation, causal analysis, and decision support;
• model deployment, monitoring, and lifecycle management; and
• data science communication, documentation, and stakeholder collaboration.

These responsibilities are fulfilled through distinct tasks and subtasks, each requiring one or more actions with specific skill requirements. Human skills and agentic skills both contribute to these responsibilities, but their levels and forms of involvement differ according to the responsibility, task, action, data, analytical context, technical environment, and expected outcome.

1.2. Human Skills and Agentic Skills in Data Science

Data science increasingly integrates both human skills and agentic skills.

Human skills are reusable capabilities possessed and developed by data scientists. They include statistical reasoning, machine learning judgment, mathematical understanding, domain interpretation, problem formulation, experimental thinking, data ethics, critical thinking, stakeholder communication, uncertainty interpretation, and professional judgment.

Agentic skills are reusable capabilities available to AI agents through their underlying models, contextual information, instructions, constraints, tools, and computational resources. In data science, these skills include information retrieval, code generation, data profiling, data cleaning support, feature transformation, statistical calculation, model comparison, hyperparameter search support, visualization generation, documentation generation, anomaly detection, summarization, and tool use.

Both skill types contribute to the same occupational responsibilities, but their involvement varies by action. Understanding a loosely defined analytical problem relies primarily on human knowledge of organizational objectives, stakeholder needs, and decision consequences; an AI agent then uses agentic skills to retrieve relevant information, profile available datasets, and suggest analytical approaches. In model development, the data scientist first determines the target variable, evaluation criteria, assumptions, and acceptable tradeoffs, while an AI agent may generate baseline code, compare model families, and organize experiment results. For communication, an AI agent may draft explanatory summaries or visualizations, while the data scientist interprets the output to ensure it represents the evidence accurately and does not overstate model reliability. Across these actions, human and agentic skills are activated sequentially or iteratively based on task requirements, with the human professional maintaining oversight and final judgment.

The relationship between these skills is not a fixed division where certain responsibilities permanently belong to humans and others to AI. Instead, human and agentic involvement shifts across tasks and between different actions within the same task.

1.3. Dynamic Human–Agentic Skill Integration and Collaboration in Data Science

Data science functions can be understood as an occupation where responsibilities are fulfilled through tasks and subtasks, tasks and subtasks are executed through actions, and actions require appropriate configurations of human and agentic skills.

An occupation provides the overall professional context. For data science, this context encompasses defining analytical problems, preparing and examining data, developing statistical and machine learning models, evaluating model performance, communicating analytical evidence, and supporting decisions.

A responsibility represents an expected area of professional work or an outcome that the data scientist accomplishes or maintains. Responsibilities such as data preparation, exploratory analysis, model development, model evaluation, experimentation, deployment support, and communication each contain multiple tasks.

A task is a context-dependent unit of work used to fulfill part or all of a responsibility. When a task is too broad or complex for reliable execution, it is decomposed into smaller subtasks.

Tasks and subtasks are executed through one or more actions. Actions represent reusable and configurable operations such as retrieving data, profiling a dataset, cleaning a variable, engineering a feature, fitting a model, calculating an evaluation metric, comparing model outputs, generating a visualization, validating an assumption, documenting a pipeline, or summarizing analytical findings.

Each action has skill requirements. These requirements describe the capabilities needed for effective and reliable execution. Depending on the action and its context, human skills, agentic skills, or an integration of both can satisfy these requirements.

The skills available for data science reside in a skill space containing both human skills and agentic skills. Relevant skills in this space can serve as candidate skills for executing the actions required by a particular task. A human skill in causal reasoning, for example, can serve as a candidate for determining whether an observed association can support a decision, while an agentic skill in model comparison can serve as a candidate for testing alternative predictive approaches.

Candidate skills are evaluated relative to the requirements of the action and its occupational and task context. Appropriate skills are then selected and organized into a skill configuration for execution. Depending on these requirements, a configuration may rely mainly on human skills, mainly on agentic skills, or on an integrated arrangement of both.

1.3.1. Skills Competition and Skills Collaboration

Two important relationships can occur between human skills and agentic skills when constructing a skill configuration: skills competition and skills collaboration.

Skills competition occurs when both a human skill and an agentic skill can satisfy the same or similar requirements for an action, but simultaneous application of both is unnecessary. Under a particular occupational and task context, one skill type can provide a better fit for the action than the other.

For example, both a data scientist and an AI agent can generate a standard summary of missing values, distributions, correlations, and outliers in a dataset. When the dataset structure, analytical objective, and reporting template are clearly specified, the agentic skill provides fast and scalable execution. When the summary requires interpretation of unusual data-generating conditions, ambiguous variable definitions, possible proxy variables, or decision-sensitive limitations, the data scientist’s human skills provide the necessary contextual understanding.

The purpose of skills competition is not to establish whether human or agentic skills are generally superior. Rather, it identifies which available skill provides the better fit for a particular action under particular conditions.

Skills collaboration occurs when one skill type alone cannot effectively satisfy the requirements of a task or its actions, or when combining complementary human and agentic skills produces stronger task performance.

Collaboration does not require human and agentic skills to operate simultaneously. Instead, they are activated at different stages of action execution in a sequential, iterative, or recursive manner according to evolving task requirements.

For example, a data scientist first identifies that a predictive model performs poorly for a particular subgroup and determines which aspects of the result require investigation. An AI agent then retrieves relevant model diagnostics, compares feature distributions, analyzes error patterns, and identifies possible performance differences. The data scientist subsequently interprets these patterns within the analytical and organizational context and formulates possible explanations. Based on the data scientist’s direction, the agent next performs additional validation, sensitivity analysis, or segmentation. Finally, the data scientist evaluates the resulting evidence and determines which findings are sufficiently supported and relevant to communicate to decision-makers. Throughout this sequence, human and agentic skills are activated iteratively as task requirements evolve, while the data scientist directs the investigation and evaluates its outcomes.

This dynamic relationship allows data scientists to leverage the complementary strengths of both skill types rather than relying on a permanent allocation of work.

1.3.2. Skill Configuration and Task Execution

A skill configuration specifies which human skills and agentic skills are used to execute the actions associated with a task or subtask and how those skills are coordinated.

The configuration depends on factors such as action requirements, data availability, data quality, problem ambiguity, statistical complexity, model risk, expected reliability, available tools, information sensitivity, decision consequences, stakeholder expectations, computational constraints, and intermediate results.

For one action, an agentic skill may independently satisfy the requirements. For another, a human skill may provide the better fit. A complex task may require several configurations across multiple actions, with human and agentic skills repeatedly interacting as the task progresses.

The configuration is therefore dynamic rather than permanent. It may change when the action changes, new information becomes available, an agentic result fails to satisfy expected requirements, or human interpretation identifies a new direction for the analysis.

Throughout this process, the data scientist remains the professional responsible for directing and controlling the work associated with the occupation and its responsibilities. Agentic AI provides powerful complementary capabilities that expand what the professional can retrieve, clean, transform, calculate, model, compare, visualize, summarize, document, and communicate.

1.3.3. Task Performance Evaluation and Skill Space Adaptation

The performance produced through a selected skill configuration is evaluated against predefined evaluation metrics. These metrics determine whether the human skills, agentic skills, or their integrated configuration are sufficiently capable of performing the relevant task and its required actions.

Evaluation criteria depend on the work being performed. In data science, they include analytical accuracy, statistical validity, data consistency, calculation correctness, reproducibility, model performance, interpretability, fairness, robustness, technical reliability, reporting clarity, relevance to stakeholder requirements, and other task-specific expectations.

When a selected skill or skill configuration produces performance above the defined threshold, it passes the performance evaluation and is retained for future applications under appropriate conditions. Repeated successful use provides additional evidence about where that skill or configuration performs effectively.

When performance falls below the required threshold, the corresponding skill or configuration is not reused without modification. The relevant human or agentic skills may be refined or reconfigured, existing skills can be combined differently, or new human and agentic skills may be developed or constructed from scratch when the existing skill space lacks adequate capability.

For example, an AI agent first generates machine learning code that executes successfully and produces acceptable validation metrics, but the resulting pipeline repeatedly applies a preprocessing step that creates data leakage. Task performance evaluation identifies this methodological problem as falling below the required threshold. The agentic skill is then refined through clearer data-splitting instructions, stronger leakage-detection checks, validation procedures, or access to approved modeling standards. The revised skill is evaluated again, and once it satisfies the required performance criteria, it is retained for appropriate future modeling tasks. Human skills follow a corresponding development process: when a data scientist lacks the knowledge required to evaluate a newly introduced modeling method, the relevant capability is developed through learning, collaboration, experimentation, and practical experience before its performance is evaluated through subsequent work.

Refined and newly constructed skills that satisfy the required performance criteria are added to the skill space. In this way, task performance evaluation provides feedback not only for completing current work but also for continuously optimizing and expanding the available skill space.

The skill space is not constructed as an isolated collection of occupation-specific representations. Many human and agentic skills used in data science are reusable across occupations. Information retrieval, data analysis, statistical reasoning, calculation, coding, summarization, visualization, communication, technical comparison, documentation, and tool use contribute to work in business intelligence, software engineering, operations research, financial analysis, marketing analytics, scientific research, and many other fields.

Reusable skills can therefore be represented once and configured differently according to the responsibilities, tasks, actions, and contexts of different occupations. This allows the skill space to serve a growing range of occupations without unnecessarily duplicating or exploding the number of skill representations. Occupational specialization is introduced through task context, action requirements, parameters, data sources, tools, constraints, and skill configurations rather than by creating a completely separate skill representation whenever the same underlying capability appears in another occupation.

The resulting process is adaptive: occupational responsibility → task and subtask → actions → skill requirements → candidate human and agentic skills → skill configuration → task execution → task performance evaluation → skill refinement, reconfiguration, or construction → optimized and expanded skill space

This cycle allows human and agentic capabilities to evolve alongside the requirements of data science work.

1.4. Essential Responsibilities of Data Scientist

1.4.1. Analytical Problem Framing and Data Science Requirement Definition

Analytical problem framing and data science requirement definition establishes the purpose, scope, assumptions, and decision relevance of data science work. This responsibility includes translating stakeholder questions into analytical objectives, defining target outcomes, identifying constraints, specifying evaluation criteria, assessing feasibility, and determining whether a proposed problem is suitable for descriptive analysis, predictive modeling, experimentation, optimization, or another analytical approach.

Agentic skills contribute extensively to information retrieval, requirement organization, method suggestion, prior-work summarization, assumption listing, and analytical planning support. Human skills remain essential for determining whether the problem is meaningful, understanding the organizational consequences of the analysis, resolving ambiguous objectives, identifying ethical concerns, and deciding which analytical formulation appropriately supports the intended decision.

For example, a stakeholder first requests a model to predict an operational outcome. The data scientist clarifies the decision the model is expected to support, determines whether prediction is the appropriate analytical framing, and identifies the consequences of false positives and false negatives. An AI agent then organizes available requirements, retrieves related analytical approaches, lists possible data sources, and proposes candidate evaluation metrics. The data scientist reviews these suggestions, rejects metrics that would encourage misleading conclusions, and defines an approved problem statement. Based on this direction, the agent prepares a structured analytical plan. Finally, the data scientist verifies that the plan aligns with the decision context and is suitable for subsequent data science work.

1.4.2. Data Acquisition, Preparation, and Feature Engineering

Data acquisition, preparation, and feature engineering establishes the data foundation required for analysis and modeling. This responsibility includes identifying relevant data sources, retrieving authorized data, assessing data quality, cleaning records, transforming variables, joining datasets, creating features, managing missing values, and preparing reproducible data pipelines.

Agentic skills are well suited to data profiling, code generation, data cleaning support, schema comparison, transformation suggestions, missing-value analysis, and repeated pipeline operations. Human skills contribute to determining which data appropriately represents the analytical question, understanding variable meaning and limitations, recognizing measurement bias, defining valid transformations, and deciding whether engineered features are conceptually justified.

For example, a data scientist first defines the data requirements for a model intended to estimate future behavior. An AI agent then profiles available datasets, identifies missing values, detects inconsistent formats, proposes join keys, and generates initial data preparation code. The data scientist reviews these outputs and identifies that one proposed feature may indirectly encode information unavailable at prediction time. The agent then revises the pipeline according to the corrected feature definition and executes consistency checks. Finally, the data scientist verifies that the prepared dataset is analytically valid and reproducible. This sequence combines human interpretation of data meaning and modeling constraints with agentic data inspection, transformation, code generation, and verification.

1.4.3. Exploratory Data Analysis and Statistical Pattern Identification

Exploratory data analysis and statistical pattern identification transforms prepared data into initial evidence about distributions, relationships, segments, anomalies, trends, and possible explanatory factors. This responsibility includes calculating descriptive statistics, visualizing data, comparing groups, identifying irregularities, examining relationships, and forming hypotheses for further analysis.

Agentic skills support repeated calculations, visualization generation, correlation analysis, anomaly detection, clustering support, pattern summarization, and rapid exploration across many variables. Human skills are essential for determining meaningful analytical questions, recognizing spurious relationships, interpreting patterns within context, understanding uncertainty, and deciding which findings warrant deeper investigation.

For example, an AI agent first identifies a strong association between a behavioral variable and an outcome measure during exploratory analysis. The data scientist reviews the association and considers whether it may reflect timing, sampling, confounding, or measurement effects. The agent then performs requested stratified comparisons, distribution checks, and visualization alternatives. The data scientist evaluates the resulting evidence and identifies which patterns appear stable and which may be artifacts of the dataset. The agent summarizes the confirmed exploratory findings and documents unresolved questions. Human and agentic skills interact iteratively as the analysis moves from broad pattern detection to context-aware interpretation.

1.4.4. Machine Learning Model Development and Optimization

Machine learning model development and optimization creates predictive, classification, ranking, recommendation, forecasting, or generative models that address defined analytical objectives. This responsibility includes selecting model families, building baseline models, training candidate models, tuning hyperparameters, managing cross-validation, comparing approaches, and improving performance under defined constraints.

Agentic skills assist with code generation, model implementation, hyperparameter search support, automated comparison, experiment tracking, documentation, and diagnostic summarization. Human skills are especially important for selecting appropriate modeling strategies, recognizing mismatch between model behavior and problem requirements, defining acceptable tradeoffs, controlling overfitting, and ensuring that optimization does not undermine interpretability, fairness, or operational usefulness.

For example, a data scientist first determines that a classification model must balance predictive accuracy with interpretability and stable performance across subgroups. An AI agent then generates baseline model code, runs candidate algorithms, summarizes validation scores, and proposes tuning ranges. The data scientist reviews the results and identifies that a higher-performing model is too difficult to justify for the intended decision context. After the data scientist revises the modeling constraints, the agent executes additional training runs and compares interpretable alternatives. Finally, the data scientist selects the model configuration that best satisfies the approved analytical, technical, and decision requirements.

1.4.5. Model Validation, Testing, and Performance Evaluation

Model validation, testing, and performance evaluation determines whether analytical models are reliable, valid, robust, interpretable, and appropriate for their intended use. This responsibility includes selecting evaluation metrics, testing model performance, examining error distributions, checking subgroup behavior, assessing calibration, detecting leakage, evaluating robustness, and documenting limitations.

Agentic skills contribute to metric calculation, test generation, error analysis, fairness checks, robustness testing, model explanation support, visualization preparation, and validation report drafting. Human skills remain essential for defining what acceptable performance means, interpreting tradeoffs, identifying methodological weaknesses, evaluating decision consequences, and determining whether model performance is sufficient for use.

For example, an AI agent first generates a validation report showing that a model meets the primary performance metric. The data scientist reviews the report and identifies that the metric alone does not capture the cost of errors in a sensitive subgroup. The scientist specifies additional subgroup tests, calibration checks, and threshold comparisons. The agent executes the requested validation procedures and visualizes performance differences. The data scientist then evaluates whether those differences are acceptable, whether model adjustments are required, and how limitations should be communicated. The final validation judgment remains under human control because the acceptability of performance depends on context, consequences, and stakeholder expectations.

1.4.6. Experimentation, Causal Analysis, and Decision Support

Experimentation, causal analysis, and decision support evaluates whether observed relationships can inform interventions, policy choices, product changes, operational decisions, or strategic actions. This responsibility includes designing experiments, defining treatment and control conditions, assessing causal assumptions, analyzing experimental results, comparing scenarios, estimating effects, and translating evidence into decision-relevant guidance.

Agentic skills retrieve methodological references, generate analysis code, calculate effect estimates, check assumptions, summarize experimental results, compare scenarios, and organize evidence. Human skills evaluate whether causal assumptions are plausible, determine whether an experiment is ethically and operationally appropriate, interpret effect sizes, resolve conflicting objectives, and decide which findings are sufficiently justified for action.

For example, an organization may want to know whether a proposed change will improve a target outcome. The data scientist defines the decision context, identifies the causal question, evaluates whether experimentation is feasible, and specifies the required comparison. An AI agent then drafts experiment analysis code, calculates sample-size estimates, organizes historical context, and proposes possible validity threats. The data scientist reviews these outputs and identifies a confounding factor that must be controlled. The agent performs additional comparisons under revised assumptions. Finally, the data scientist synthesizes the evidence and communicates what the results may support, what they do not prove, and which decisions remain uncertain.

1.4.7. Model Deployment, Monitoring, and Lifecycle Management

Model deployment, monitoring, and lifecycle management supports the reliable use of data science outputs in production, analytical, or decision-support environments. This responsibility includes preparing model artifacts, supporting integration, defining monitoring metrics, detecting data drift, tracking model performance, managing retraining requirements, documenting versions, and coordinating updates with technical and business stakeholders.

Agentic skills assist with deployment script generation, monitoring logic, drift detection, alert summarization, documentation, dependency inspection, reproducibility checks, and repeated maintenance actions. Human skills remain important for defining operational requirements, understanding consequences of model failure, interpreting drift, approving retraining decisions, controlling changes, and ensuring that deployed models continue to serve their intended purpose.

For example, a deployed model begins producing unusually different score distributions. An AI agent detects the change, compares input feature distributions, summarizes possible drift sources, and generates diagnostic plots. The data scientist reviews the diagnostics and determines whether the shift reflects a data pipeline issue, a real change in the underlying population, or an expected seasonal pattern. The agent then performs additional checks requested by the data scientist and prepares a retraining impact summary. Finally, the data scientist decides whether to continue monitoring, recalibrate the model, retrain it, or recommend temporary suspension. This collaboration allows technical monitoring to scale while keeping consequential lifecycle decisions under professional judgment.

1.4.8. Data Science Communication, Documentation, and Stakeholder Collaboration

Data science communication, documentation, and stakeholder collaboration converts technical work into understandable evidence, recommendations, limitations, and reusable knowledge. This responsibility includes explaining methods, presenting results, documenting assumptions, creating visualizations, communicating uncertainty, responding to stakeholder questions, and ensuring that analytical outputs can be understood and reused responsibly.

Agentic skills summarize analyses, draft documentation, generate visualizations, adapt explanations for different audiences, organize evidence, prepare presentation materials, and produce alternative wording. Human skills evaluate the meaning of results, identify what stakeholders need to understand, avoid overclaiming, communicate uncertainty appropriately, resolve conflicting interpretations, and determine which recommendations are justified.

For example, an AI agent first drafts a model results summary with performance metrics, feature importance charts, and recommended actions. The data scientist reviews the draft and identifies that the explanation overstates causality and underemphasizes a data limitation. The data scientist provides corrective instructions and specifies how uncertainty should be communicated. The agent revises the narrative, updates the visuals, and prepares audience-specific versions. Finally, the data scientist verifies that the final communication is accurate, decision-relevant, transparent, and appropriate for the stakeholder group.

1.5. Changing Human and Agentic Involvement Across Data Science Responsibilities

Human skills and agentic skills are involved throughout data science, but their relative involvement differs across responsibilities and changes during task execution.

These involvement levels are not permanent allocations. A single responsibility may contain actions where human and agentic skills compete as alternative capabilities and other actions where they collaborate sequentially, iteratively, or recursively.

The practical objective is to construct skill configurations that use the most suitable available capabilities for each part of the work and revise those configurations when task performance evaluation shows that another arrangement performs more effectively or reliably.

1.6. Working as a Data Scientist with Agentic AI

Agentic AI is becoming an increasingly powerful component of data science. Data scientists should therefore adopt and integrate AI agents as complementary participants in professional workflows rather than treating them only as occasional tools for generating code, summaries, plots, or reports.

This shift does not reduce the importance of data science expertise. Human professionals remain responsible for the occupation and its corresponding responsibilities. They understand analytical objectives, define meaningful questions, interpret data context, determine appropriate methods, evaluate model validity, communicate uncertainty, identify ethical and operational risks, control consequential decisions, evaluate task performance, and decide how human and agentic skills are configured and improved.

At the same time, agentic skills greatly expand the volume and variety of analytical work that can be processed. AI agents retrieve and organize large volumes of information, generate and test code, profile datasets, create features, compare models, calculate metrics, detect anomalies, prepare visualizations, draft documentation, monitor deployed systems, and synthesize extensive analytical evidence at a scale difficult to achieve through direct human effort alone.

Strong data science practice is therefore not based on maximizing human work or agentic work independently. It is based on finding appropriate configurations of human skills and agentic skills for the responsibilities, tasks, subtasks, and actions being performed.

Where a human skill and an agentic skill can independently perform the same action, skills competition identifies which is better suited to the current context. Where work requires complementary capabilities, skills collaboration results in human and agentic skills being activated sequentially, iteratively, or recursively to achieve stronger task performance than either skill type achieves alone.

Task performance evaluation then determines whether selected skills and configurations satisfy predefined expectations. Successful skills are retained and reused, while inadequate skills are refined, reconfigured, or newly developed. The resulting capabilities return to an increasingly useful skill space that supports data science and, where capabilities are reusable, other occupations as well.

For a data scientist, working effectively with agentic AI means treating AI as a powerful complementary capability integrated into everyday professional routines while maintaining human control of occupational responsibilities. Through appropriate skill selection, competition, collaboration, evaluation, and continuous improvement, human professionals and agentic AI work together to achieve data science work that is more efficient, scalable, adaptable, timely, rigorous, and reliable.