In this week’s edition of *The AI Shift*, Financial Times data reporters John Burn-Murdoch and Clara Murray discuss how AI agents are transforming their workflow. Clara Murray, initially skeptical of AI tools, experimented with OpenAI’s Codex platform for two major data projects: scraping multilingual product prices and building an interactive calculator from millions of data points. While Codex enabled rapid prototyping—producing a calculator without designer or developer input and scraping thousands of prices—it also introduced significant challenges.
The system operated with minimal transparency, requiring constant supervision. Murray found herself glued to her screen late into the night, managing dozens of parallel agents. Her laptop crashed under the load, and she discovered critical errors: one agent guessed prices by selecting the largest number on a page, while another confused color variants.
Without thorough input verification, the calculator’s outputs were unreliable. The experience highlighted a paradox: AI agents accelerate creation but complicate validation. Princeton professor Arvind Narayanan described managing agents as combining the worst aspects of leadership—increased stress and accountability—without the benefits.
For data journalists, where accuracy is paramount, the tools offer speed but demand rigorous oversight. Murray continues using Codex but with heightened awareness of its limitations. Burn-Murdoch, who has managed agent swarms longer, is invited to share his insights on where the technology delivers the most and least value.
Source: Financial Times Companies · Summarized by HeadlinesBriefing