IBM Plunges After Anthropic's Latest Update Takes on COBOL

Original Article ↗ Hacker News Discussion ↗

Why COBOL/Mainframes Persist

  • Banks and large institutions stay on COBOL mainly due to massive, battle‑tested codebases and the stability of mainframes, not because COBOL itself is hard.
  • Mainframe architectures (e.g., sysplex) are praised for extreme reliability, virtualization, and hardware abstraction; outages are rare relative to scale.
  • Licensing is expensive, but migrations are riskier: decades of real‑world behavior, regulations, and manual test requirements create huge inertia.
  • Some note banks are slowly moving to distributed/event‑based systems, driven by cost and competition from neo‑banks, but progress is slow and fraught.

Anthropic’s COBOL Pitch and LLM Capabilities

  • Anthropic’s claim is framed less as “we write COBOL” and more as “we analyze your COBOL and generate a migration plan/target code.”
  • Optimists see value in models that can ingest entire legacy codebases (plus history) and help humans navigate, document, and gradually port them.
  • Some suggest LLMs could assist in reverse‑engineering lost binaries and easing modernization, especially for peripheral jobs/batch tools.

Skepticism: Safety, Logic, and Training Data

  • Many doubt LLMs can safely untangle 50+ years of “spaghetti” in mission‑critical finance, insurance, and rail systems.
  • Concerns center on small public COBOL corpora, hallucinations, and “shotgun surgeon” edits causing billion‑dollar failures.
  • Several argue the real risk is management using AI hype to justify reckless changes and underpaying/retiring experts, setting up a payments‑infra crisis.
  • Some insist no serious CIO will let a chatbot rewrite core banking logic; at best, AI assists humans, and migrations require long parallel runs.

Business Logic vs. Language

  • Repeated theme: the hard part isn’t COBOL syntax but the embedded business rules, regulatory quirks, and historical context.
  • Converting COBOL to Python/Go/.NET doesn’t remove complexity; it becomes a dangerous full rewrite. Prior tools (e.g., COBOL‑to‑x86/.NET) never truly disrupted IBM.

Impact on IBM, Oracle, and AI Bubble

  • Debate over whether AI‑assisted migration is an existential threat to IBM’s mainframe business or overblown panic in an overheated AI market.
  • Some see stock drops as speculative churn (sell, scare, buy the dip), noting IBM mainframe growth and prior limited success of past “modernization” vendors.
  • Oracle is mentioned as also weak on frontier models, though its stock reaction is less discussed.

Source and Meta Discussion

  • Multiple commenters criticize the ZeroHedge article as low‑quality and politically toxic, questioning why it’s on HN at all.