“A man who carries a cat by the tail learns something he can learn in no other way.”
– Mark Twain

‍

One area in which the generative AI hype is falling back to earth is its applicability to mainframe modernization. The swift, emphatic replies to the Anthropic blog about Claude converting COBOL (yep, the one that temporarilywiped $30 billion off IBM’s market cap) surfaced just how darn hard migrating off a mainframe actually is. From SVPs at IBM to our own CEO, the consensus landed in the same place: migration is more than code translation.

So now, eminent pundits like Gartner are counseling a retreat¹: The mainframe isn’t that bad after all. IBM is doing a great job keeping it relevant. You should “modernize in place” rather than working to get off the mainframe entirely. For large shops, a full exit is “nearly impossible.” For medium-sized environments, 40% should simply do nothing.

And that is all true.

And yet.

‍

There are many cases in which good enough is, well, good enough. But there are other instances in which good enough is just kicking the can down the road. Do you really want to build real-time fraud detection on a system that wasn’t designed for millisecond decisions? Or deliver the personalized, AI-driven customer experience your competitors are already shipping, on infrastructure that predates the internet?

The notion that AI makes migration trivially easy has done genuine damage. It’s setting up organizations to get burned yet again by a shiny new technology, and crucially, permitting many intelligent people to conclude that the whole enterprise is hopeless. But the hype is stuck in a rut: apply generative AI as an accelerant to the same approaches that have always struggled, a failure pattern that we’ve noted over and over again.

Consider what AI has done elsewhere. In drug discovery, screening billions of molecular candidates was always the ideal, but impossibly expensive. In radiology, reading every scan rather than triaging by eye was always better, but impractical at volume. AI made the right process viable. That same shift unlocks pragmatic, incremental, non-disuptive, risk-mitigated mainframe migration that works.

Even Gartner’s latest mainframe migration framework* rests on an assumption that is becoming less relevant: that complexity scales with codebase size in ways that make large environments categorically harder to migrate than small ones. A large, tangled codebase is still harder to migrate than a small one — but the gap has narrowed considerably, and it will keep narrowing. Segmenting organizations by MIPS and assigning each a pre-determined strategic fate doesn’t account for any of that.The right question is less how big your mainframe is, and more whether your mainframe is constraining what your organization needs to do next. If the answer is yes, retreating to safety doesn’t make the problem go away. Nor does jumping on the latest silver bullet. It just means someone else will have to deal with it — using up yet another life.

‍

‍

News and Views

At what point do hallucinations go from being slightly entertaining to disturbing?

While generative AI’s confident tone has appealed to our biases to believe “authority,” imagine the dangers of inaccuracies when mission-critical systems are at stake.

If you’re worried about the tech debt you currently have, AI slop can be a nightmare. Gergely Orosz and Elin Nilsson of the Pragmatic Engineer wrote in a recent post about the three engineer archetypes (builders, shippers, and coasters), and how the shippers actually might accumulate tech debt faster and risk building the wrong things.

Relevant op-ed from The Economist that companies are "de-weirding" AI — defaulting to automation over imagination — and in doing so are squandering what makes it transformative. The same trap applies to mainframe modernisation: those who simply apply AI to broken approaches like code translation are de-weirding the technology before it even has a chance to matter.

‍

‍

From the Orchard

We’re at Google Next this week! Hear directly from a customer of ours², or attend this panel on AI’s role in transforming operations. And of course, we’d welcome the opportunity to meet.

Making agents dependable requires more than “prompt tricks”. On our Substack, Jeff Schomay breaks down this emerging practice — and also goes one step further in explaining just how to orchestrate confidence when you’re not even writing the prompts yourself anymore.

Through Carahsoft’s GSA Schedule Contract, our mainframe modernization platform Imogen is now available for procurement to educational institutions and governments nationwide: federal, state, and local. If you missed out on the Carahsoft-hosted webinar we took part in last week, you can still request to watch it here.


Curious to learn more? Say hello@mechanical-orchard.com.‍

‍

—

View all newsletters here

‍