Data Workers Lag Behind Developers in Adopting Agentic AI Tools
Business analysts struggle to trust AI agents because outputs lack transparency and source data is often outdated or incomplete.
Key points
- US business roles outnumber developers six to one but see fewer AI productivity gains.
- Data workers cannot verify if agents used correct formulas or complete datasets.
- Reliance on stale data copies creates shadow versions of truth that erode trust.
Business and financial operations professionals in the United States are falling behind software developers in realizing productivity gains from agentic artificial intelligence tools, according to a report by InfoWorld. While engineers have rapidly integrated these agents into their workflows, data-centric workers remain hesitant due to significant trust and verification barriers that prevent widespread adoption.
The primary obstacle for data professionals is the inability to easily validate the results produced by AI agents. Unlike code, which developers can check by reading diffs and running tests, AI outputs for analysts often appear as final numbers, charts, or opaque blocks of code. This lack of transparency means users cannot confirm whether the agent applied the correct formulas, combined data accurately, or missed critical information during the analysis process.
The Transparency Gap in AI Outputs
Programmers adopted agentic AI quickly because the technology produced outputs in a format they already understood and could rigorously test. When an agent modifies a function, the developer can inspect the changes and run existing tests to decide whether to accept the work. This process requires minimal faith in the machine because the evidence of correctness is visible and verifiable through standard engineering practices.
Data analysts, however, face a black-box scenario where the reasoning behind a result is hidden. An analyst might request a specific analysis, but the return value offers no insight into the methodology used to generate it. Without a clear audit trail showing which data points were included or how calculations were performed, the output cannot be automatically assumed to be accurate or reliable for business decisions.
Data Access and Lineage Issues
A major factor contributing to this distrust is the difficulty business users face in accessing current, authoritative data. Relevant information often resides in disparate systems such as Salesforce, Databricks, Snowflake, or BigQuery, requiring formal requests to data platform teams. These teams are frequently understaffed, leading to delays and workarounds that involve using outdated copies of data rather than live sources.
Consequently, analysts often work with shadow versions of the truth, such as CSV files pulled months ago or spreadsheets passed through multiple hands with untracked changes. When an AI agent processes this stale or fragmented data, the resulting analysis lacks proper lineage. In financial institutions, where accuracy is critical, relying on such opaque processes based on questionable data sources poses significant professional risks.
What should you do?
Organizations should recognize that current agentic AI tools may not be ready for unsupervised use in critical data analysis roles. Until tools provide transparent audit trails and direct access to governed data sources, human validation remains essential. Security and operations teams should monitor how these agents interact with sensitive data repositories to prevent the propagation of errors based on outdated information.
What to do and how to stay safe: agentic AI
- Verify that AI-generated data analyses include detailed audit trails showing source data and calculation methods.
- Ensure business users access live, governed data platforms rather than stale local copies or spreadsheets.
- Implement manual review processes for AI outputs before using them for financial or operational decisions.
- Train data platform teams to handle increased request volumes to reduce reliance on outdated data workarounds.
Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Why are data workers slower to adopt AI than developers?
Data workers cannot easily verify if AI agents used correct data or formulas, whereas developers can test code outputs.
What is the main barrier to trusting AI in data analysis?
The output is often a black box that does not show how results were produced or if data was complete.
How does data access impact AI reliability for business users?
Users often rely on outdated copies of data due to understaffed platform teams, leading to inaccurate AI results.
