FT Data Reporters Test AI Agents on 2 Projects, Finding Speed Gains but Accuracy Risks
Updated
Updated · Financial Times · Oct 29
FT Data Reporters Test AI Agents on 2 Projects, Finding Speed Gains but Accuracy Risks
1 articles · Updated · Financial Times · Oct 29
Summary
Two FT data reporters said AI agents sharply accelerated coding and data work, but only after heavy human oversight and verification to catch serious errors.
Clara Murray used Codex on 2 large projects and quickly produced a calculator prototype and scraped thousands of prices, yet later found agents had guessed prices from the biggest number on pages and mishandled product variants.
John Burn-Murdoch said the payoff comes when agents handle routine, checkable tasks; he now builds audit trails, runs AI-led error checks and sometimes rewrites analyses from scratch to confirm striking results.
For data journalism, where accuracy is critical, both reporters said agentic AI can expand what is possible but also turns journalists into managers of nonstop outputs, making independent thinking and final editorial judgment more important.