How Morgan Stanley is using its DevGen.AI tool, built in-house on OpenAI's AI models, to translate legacy code into English specs for developers to then rewrite
Morgan Stanley is now aiming artificial intelligence …
Context & Ripple Effects
Morgan Stanley’s use of an internal tool built on OpenAI models turns legacy-code interpretation into a structured handoff: the model produces English specifications, while developers retain responsibility for the rewrite. That places AI first in the documentation and comprehension layer of software modernization rather than presenting it as an autonomous coding replacement.
The story extends the shift from broad employee experimentation with generative AI to institution-specific workflows; related coverage also shows a peer bank planning to augment its workforce with an AI software engineer.
First-order effects
- Morgan Stanley developers can use DevGen.AI to extract readable specifications from hard-to-interpret legacy code before rewriting it, changing the starting point of modernization work.
- OpenAI’s models become a component inside a bank-controlled development workflow, while human developers remain the implementation and validation layer.
Second-order effects
- Other financial firms pursuing comparable legacy-system work face added pressure to test AI-assisted code understanding, especially as Goldman Sachs moves toward AI augmentation for software-engineering work.
- The near-term value shifts toward workflow design, review controls, and the quality of generated specifications—not merely access to a general-purpose model.
Third-order effects
- If this pattern holds, enterprise AI adoption will increasingly be measured by whether firms can embed models in bounded, auditable processes around their proprietary systems, rather than by standalone chatbot usage.
- Legacy-code modernization may become an early proving ground for a broader division of labor: models translate and summarize system knowledge, while accountable staff make production changes.
The trend: Frontier models are being institutionalized as controlled workflow components that help enterprises convert accumulated internal knowledge into operational change.