Morgan Stanley's AI tool reviewed 9M lines of COBOL, saved 280K hours as GPT-6.1 Sol hits 98% on migration eval
SumitGup · x · 2026-10-22
Morgan Stanley built DevGen.AI in-house on OpenAI's GPT models to translate legacy COBOL into plain-English specs. Per global tech chief Mike Pizzi, it reviewed 9 million lines of code and saved developers 280,000 hours in H1 2025.
Context: COBOL, now 67 years old, still powers 40%+ of online banking, 80% of in-person credit card transactions and 95% of ATM transactions.
OpenAI meanwhile reports GPT-6.1 Sol scores 98% on its COBOL-to-modern-language migration eval, where a rewrite only passes if every output matches the original byte for byte — one wrong rounding rule could corrupt everyone's balances.
More from coding & agent
- Richard Socher's Recursive launches Flint, a local coding model built in 4 months by the Eureka Machine — aidenybai · 2026-10-22
- Ropedia's S-Agent 2.0 Tops Spatial Reasoning Benchmarks With Zero Training, Scoring 66.5 on MMSI — CyberRobooo · 2026-10-22
- Eden Deploys Vercel's Eve Agent to Cloudflare With One Command, Fully Open Source — ritakozlov · 2026-10-22
- Ben Bajarin: rambling research notes into AI for organized workshop analysis works well — BenBajarin · 2026-10-22
- Reflection ships open Beam customization recipes spanning LoRA, SFT and GRPO RL — HowDevelop · 2026-10-22
- Hooked a cheap receipt printer to Claude for proactive briefings, notifications and even bird profiles — beglen · 2026-10-22