AI Agents and Analytics: Responsibility Remains Human

AI agents understand metrics, identify patterns, write analyses, and can even execute certain actions. However, a practitioner in analytics reminds us that the responsibility for choices, prioritization of topics, and the formation of an initial viewpoint remain deeply human. She details how to delegate without relinquishing judgment.

The Advantage Shifts to Judgment and Aspiration

As artificial intelligence takes on an increasing share of execution tasks, human added value is becoming more focused on judgment, aspiration, and the ability to think independently. It is advisable to invest in human skills that sharpen with experience, such as critical thinking, creativity, discernment, leadership, and vision. The coming months may present an opportunity to transform acquired experience into frameworks, principles, and reusable decision models, while exploring opportunities at the intersection of strategy, analytics, and AI, all while staying attuned to advancements in the field.

What AI is Already Doing in the Analytics Stack

Current AI can generate SQL and summarize dashboards. Equipped with semantic layers, agents understand metrics, detect patterns, recommend actions, and, in some cases, autonomously execute these actions by integrating business rules, past decisions, and enterprise risks. They now cover query writing, documentation, reporting, analysis, research, and narrative formulation of recommendations. In this sense, the analytics stack is described as being rewritten, with AI demonstrating greater capability in both analysis and execution.

The Layer the Agent Does Not Touch: Responsibility and Rule-Bending

The challenge is not so much the AI's understanding of the business but the assumption of the consequences of decisions. Deciding when to break a rule or deviate from a data-driven recommendation falls under human judgment, according to the author. Humans sometimes make decisions contrary to numerical signals due to strategic priorities, political dynamics, cultural nuances, regulatory risks, or a conviction about the path to take, and someone must be accountable for that. Even as the analytics stack changes, responsibility does not change and remains beyond the reach of agentic AI.

Risk of Diluting the Initial Perspective

The growing reliance on assistants like Copilot to interpret upstream patterns can save time, but it exposes the risk of skipping the construction of one's own initial viewpoint. The concern expressed is not primarily about the agent's error, but the ease of substituting an external interpretation for one's own thinking, risking a loss of originality and intellectual friction, which are sources of creativity and conviction. The question posed by the time saved through AI is whether it serves deeper reflection and the resolution of more ambitious problems, in a context where AI is not meant to replace human jobs but to accelerate execution.

What Really Matters for Analysts

With over seven years of experience in analytics and AI, the author observes that analysts' difficulties stem less from SQL or Python than from translating insights into decisions. She reminds us that AI does not define aspirations, does not make tough calls, does not establish trust, nor does it take responsibility for outcomes. In an environment where agentic AI absorbs execution tasks, the scarcity lies in original thinking, business judgment, and determining what deserves to be prioritized.

Delegation Lines: This Quarter, By Year-End, and Beyond

The career question posed is what to delegate to AI and what to keep under control. In the short term, it is suggested to audit the use of one's weekly time, separate mechanical tasks from judgment work, and then broadly delegate to AI the summarization, documentation, data preparation, initial drafts, and exploratory queries, while reserving energy for the questions that matter. Before engaging AI, it is recommended to take five minutes to establish one's own hypothesis and use agents to challenge one's thinking rather than substitute it. By the end of the year, the goal is to learn to work with agents as one would with a new colleague, paying particular attention to context, constraints, objectives, and feedback, and to use AI as a coach to test reasoning, detect blind spots, simulate stakeholder reactions, and consider alternative scenarios. It is also advisable to practice converting complex results into decision-making choices as technical realization becomes standardized.