“Old and new make the warp and woof of every moment. There is no thread that is not a twist of these two strands.”
— Ralph Waldo Emerson

‍

The primary reason modernization efforts are big, scary projects is because we humans have a terrible habit of deferring maintenance until something is about to break — or breaks. We’ve written before about how, in an ideal world (ha), modernization should be a competency, like compliance or cybersecurity. But, alas, too many people hope and pray that they can kick the can down the road far enough so that it falls to their successors to deal with the problem.

This head-in-the-sand approach is amplified by another challenge: orchestration between the old and the new. When part of the system moves and part of it doesn’t, something has to keep the whole thing coherent. Or, to use the common analogy, how do you fix the plane while flying it? (Make sure it’s well maintained when it’s on the ground, of course, but then where’s the drama?)

This is where AI agents become interesting. Imagine an AI agent that keeps new and old systems operating in harmony, monitoring the inputs and outputs between both. It makes sure that each workload meets three types of equivalence:
‍

• Behavioral equivalence, meaning it produces the exact same output for the same inputs
• Performance equivalence, meaning it produces the exact same output in the exact same time
• Integration equivalence, meaning it ensures the modernized workload can still communicate with unmigrated upstream and downstream consumers. 

Getting to integration equivalence is the hardest part in traditional modernization approaches — one that too many vendors have pushed onto their customers. Interfaces that look simple on paper are often carrying implicit guarantees about timing, ordering, retries, error handling, and even failure modes. A field that is technically optional may, in practice, always be present. There are even systems that have relied on undocumented side effects, like a record being written twice, to signal a manual follow-on action.

The older the system, the more dense the set of tacit expectations involving things like message queues, file drops, batch schedules, shared databases, and the random, one-off point-to-point connection. Preserving integration equivalence means faithfully reproducing all of those idiosyncrasies, even when the workload spans across old and new infrastructure.

Figure out orchestration, and the world is your oyster. It means you can pick and choose which parts to modernize, and when, based on business priorities. It means you can fit modernization into a modest budget. Choose just the workloads that need it most at that time. It effectively blurs the line between old and new entirely: every system is a glorious, comprehensible mix of legacy and modernized. Or as our partner Thoughtworks says in its newly-launched AI/works platform, “Your systems finally stop aging.”

‍

News and Views

Thoughtworks published a case study of a “leading global manufacturing company” that used Imogen, our mainframe modernization platform, to rewrite its extended warranty platform.

If you’re thinking about AI as primarily a magic optimization tool, you’re missing out. A Thoughtworks research report, Beyond the bottom line, indicates more and more company leaders are getting the message that AI will rewire existing processes and reward those who can imagine completely new ways of working.

Exhibit A why specifications are crucial: few things are more amusing than a non-deterministic LLM given an unbounded task, as this experiment from the WSJ discovered. The WSJ also published the findings of a survey that shows a stark disconnect between what CEOs think they’re gaining in productivity using AI and what their own workforces think. Hmmm.

‍

From the Orchard

We may have wound down our podcast, but we’ve started a Substack, where we’ll be featuring more pieces on how we think about technologies (especially AI) as they relate to mainframe modernization. To kick things off, Mechanical Orchard software engineer Rin Kuryloski has written about the mathematical basis for why humans-in-the-loop workflows outperform fully autonomous agents.

Is modernization reluctance a generational issue? Our COO Edward Hieatt and Thoughtworks’ CTO, Rachel Laycock, argue in Infoweek that, until companies stop attempting “big bang” rewrites, enterprise modernization will remain stalled: veteran leaders are allergic to career-ending risk, and younger engineers aren’t keen on legacy tech.

The latest “Expert Edition” from the Federal News Network covers DevSecOps for government agencies, including a contribution from Mechanical Orchard’s Edward Hieatt.

What happens when vibe coding meets mainframe administrators? Find out in our latest “The Weekly Standup” video.


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

‍

—

*Issue first published on January 28th, 2026. View all newsletters here

‍