sdecoret - stock.adobe.com
Will AI replace data analysts: A year and a half later
Organizations should foster and fund -- not terminate -- entry-level positions in favor of AI. Junior talent is key to the long-term success of AI agents.
Are data analyst jobs still safe from AI? It's complicated.
In January 2025, I argued that generative AI (GenAI) would not replace data analysts for many reasons. Since writing that original article, model capabilities have advanced considerably, and AI agents are now commonplace. What's more, AI is having a significant effect on jobs, especially in the technology field.
In May 2026, Meta dismissed roughly 8,000 people -- about a tenth of its workforce -- to fund AI spending. That same month, Cisco let go of around 4,000 employees. Amazon removed close to 30,000 corporate roles since late 2025, and Oracle cut a comparable number in March 2026. When Block cut nearly 40% of jobs, CEO and co-founder Jack Dorsey attributed them to AI tools that, he claimed, changed what it takes to run a business.
The situation is more complex than the number of jobs lost. IBM's 2025 survey of senior data and analytics executives found that 47% of respondents ranked attracting and retaining skilled people among their most difficult problems, up from 32% two years earlier. At the same time, 77% reported trouble filling data roles.
Meanwhile, the Infragistics Reveal 2026 IT Talent Survey found that among firms planning to hire, 70% directed hiring at senior professionals, particularly those with AI expertise. Only 12% planned to hire at the entry level.
Chief executives are attributing tens of thousands of job losses to AI. Research shows employers cut routine and generalist roles while trying to hire a smaller group of experienced people. In short, a firm can lay off thousands but still report that it cannot hire the analysts it wants. In this environment, everyone in the technology sector has reason to question whether their job is safe.
Improved analyses and limited insights
In 2025, I described a key constraint of GenAI: it couldn't access raw data and perform analysis directly. I also noted that AI wrote formulaic summaries and had poor visualization.
Now, AI has more features and functions. Over the last 18 months, the semantic layer -- an intermediate layer translating raw technical data into standardized, everyday business terms -- has seen remarkable growth. Using this approach, organizations can ensure both data analysts and AI models use the same definitions. An AI agent can now query governed tables directly instead of guessing from text. AI models still write tediously formulaic summaries, but visualization has greatly improved.
However, an agent still can't determine which question to ask or which answer to trust. That requires broader strategic context, critical thinking and real-world knowledge that an agent only simulates. Ultimately, AI can brilliantly automate data retrieval, but remains blind to financial, ethical and strategic consequences of its own outputs. In high-stakes business environments, human judgment is the anchor of accountability.
Take, for example, a common marketing analysis. An AI agent using the semantic layer reports that paid search drove the majority of last quarter's revenue, so it recommends a budget increase. Assuming the agent computed the figure flawlessly according to its parameters, this should be table stakes in 2026. However, a human analyst has the business intuition to look closer.
The analyst suspects that a concurrent brand campaign and a sudden price cut were the true drivers of customer demand. Consumers merely clicked a paid search ad because they were already interested, but that action triggered an attribution model that blindly gave the final click the credit. Seeing this, a human analyst can step in to correct the metric's definition within the semantic layer. The AI model lacks the real-world business understanding to do this and would continue to optimize a scenario that never really applied.
Accountability and agency
We need human beings who can bring doubt, care and ambition to their work.
Although an AI agent can write a recommendation, it can't answer for one. When an agent misprices a product, breaks a disclosure rule or includes incorrect figures in the board report, the company can't punish the model. After all, regulators require a bank or an insurer to identify the person responsible for a decision. As long as organizations assign analysis building to agents but require a person to answer for its outputs, companies will need an analyst to validate, explain and query their numbers.
Analysts now spend time on three tasks: interpreting data, writing the explanation and governing automated outputs. AI tools expertly format data, lay out reports, write baseline calculations and build charts. Once the software does the mechanical work, analysts interpret the data, find the explanation a business audience will act on and present it to those who control budgets.
Where will the new analysts come from?
In addition to shedding existing workers, companies are cutting entry-level positions, barring the next generation from the profession. A graduate learned to judge data by cleaning it, writing basic queries and drafting charts while a senior analyst reviewed their work. Agents now do the work that new entrants once learned from, so the graduate gets no such practice.
If companies still need analysts to validate AI outputs, but companies have stalled hiring, the question remains: who will do this work? And how will they acquire the necessary skills and insights? The senior analysts that every firm wants to hire in 2026 are the juniors who companies trained in 2020. A company that cuts its entry-level jobs now will lack experienced analysts in five years.
The verdict
So, a year and a half later, are data analysts safe? The answer depends on the type of analyst. An analyst who maintains the semantic layer, audits the agents and knows the business well enough to judge an AI output's accuracy, validity, ethics and relevance might keep a secure job -- chief data officers cannot hire enough such people. Meanwhile, an analyst who spends their day on routine extraction, formatting and baseline charts might lose that work to software.
I recommend two actions.
Any analyst working today should understand the semantic layer and the business well enough to judge an AI agent's performance and take responsibility for metrics and definitions.
An organization should keep its entry-level jobs, fund junior analysts' training and assign a responsible analyst to oversee and govern AI agents.
Today, software does a large part of the analyst's work: considerably more than I reported in early 2025. But it's still true that people -- and only people -- answer for that work, judge the outcomes and train the next, still-needed generation of analysts.
Donald Farmer is a data strategist with 30+ years of experience, including as a product team leader at Microsoft and Qlik. He advises global clients on data, analytics, AI and innovation strategy, with expertise spanning from tech giants to startups. He lives in an experimental woodland home near Seattle.
