Franklin Templeton taps OpenAI Codex for investment research

Franklin Templeton's early attempt at AI put data in a database and connected models to it, but the models did not understand how the data fit together, so the answers fell short. Adding agents and a semantic layer — a structured model of what the firm's data means and how it connects — changed the results, according to Dudley. That experience shaped the firm's current push to build a governed, reusable intelligence foundation that brings together trusted connectors, portfolio and workflow context, and agentic capabilities.

The firm's framing has shifted from automating a human-led, sequential equity-research workflow — where an analyst performs first-pass stock analysis, reviews earnings calls and studies competitors — to what Dudley calls continuous intelligence combined with human judgment. Franklin Templeton has used Codex in hackathons with investment practitioners and is now working to turn experiments into production-ready tools with security review, expert support and data controls.

A key design concern is that many AI tools learn a user's preferences and show more of the same, which the firm sees as a potential source of biased perspective in investing. Franklin Templeton wants the opposite — access to diversity of research and opinion — and wants its systems to emphasize investment risks, so portfolio managers are aware of them when making decisions.