“If something isn’t working, don’t try harder or do more. Do something different.”
— Ramit Sethi

‍

In December 2023, Gartner predicted that GenAI would “supercharge legacy modernization” in 2024. They were right, at least from a marketing buzz perspective.

‍

From watsonx CodeAssistant for Z to AWS Mainframe Modernization Refactor with AWS Blu Age (brevity clearly isn’t the industry’s strong suit), both legacy incumbents and cloud providers have piled into territories once ruled by the giant global system integrators.

‍

But there’s a major flaw: while GenAI has made massive strides in helping us understand legacy systems, these efforts layer onto the same, code-centric modernization approaches that have been synonymous with high failure rates over decades. Surveys and case studies have reluctantly called out the long and uncertain tail of post-transpilation efforts, which looks something like this:

‍

‍

Bolting GenAI onto a sub-optimal approach is like putting a jet engine on a Conestoga wagon. There’s a reason RFP responses from giant system integrators lovingly detail the “comprehension and translation” stages while keeping the later phases deliberately vague, both in timeline and cost.

‍

While we’ve written a lot about why current approaches aren’t successful (see “Comprehension > Code,” Kent Beck’s classic, “Confidently,” and “First, then”), we haven’t written as much about our ability to harness the power of GenAI in a way that produces trustworthy, high-quality, idiomatic code that eliminates technical debt and does exactly what you need it to do. Remedying now.

‍

Many AI generations ago, in June 2023, our CTO Roberto Ostinelli ran an experiment involving two AIs pair programming to create a basic banking functionality using test-driven development (TDD).

‍

We’ve evolved this early combination of TDD and GenAI into a “verify, then trust” model: every unit of code is continuously tested against real-world data flows, to make sure it behaves as expected before cutting over into production. This iterative validation process aligns the rewritten code with actual system behavior, producing audit-ready software validated by actual production data.

‍

In other words, we’re able to achieve deterministic verification of AI-generated code, radically “shifting left” the notoriously unreliable phase of integration and regression testing that plagues so many modernization efforts.  And that means incredible leaps in speed without sacrificing quality.

‍

Less risk, more speed, higher quality outcomes. We think that’s a pretty good definition of “supercharging legacy modernization.”

‍

News and Views

“Move fast and break things” is sexy, but you know what else is sexy? Getting your benefit checks on time. Wired and Fast Company cover the culture clash when Silicon Valley coders born in this century first encounter COBOL.

‍

Barclay’s, one of the UK’s biggest banks, went down on payday last month (employees in the UK are paid monthly), causing temporary homelessness for one unfortunate customer. It was not related to a cyber attack of malicious activity. The fallout has commenced.

‍

Speaking of banks, this is astonishing to anyone not in the banking sector: according to The Economist (subscription required) the industry globally spends more than $200bn on compliance eachyear. With DORA, that number isn’t going down.

‍

Neither fish nor fowl: Jake Saper and Rishub Nahar write incisive observations from the frontiers of the emerging AI-Enabled Services (and thank you for the name check!)

‍

From the Orchard

Wondering how to navigate the execution challenges of digital transformation? Noted banking transformation expert and author Dr. Leda Glyptis is holding an in-person masterclass in NYC on Friday, March 21st geared toward emerging executives who need to operationalize bold strategies. Spaces are very limited; apply here.

‍

Mechanical Orchard delivered two presentations at Rise8’s Prodacity 2025 Summit in Nashville: Nick Keune broke down the elements of how we bring confidence to modernization, and Edward Hieatt refreshed perspectives on Continuous Delivery.

‍

Our latest On a Limb podcast episode features retail technology analyst Paula Rosenblum discussing the challenges of modernization in an industry known for its razor-thin margins. Listen to the episode here.

‍

One of our customers is an F500 retailer modernizing its mainframe safely, reliably, and increasingly quickly. We’ll be at Shoptalk 2025 if you’d like to meet up with us and find out how we’re doing it.

‍

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

‍

——————

Issue first published on February 25th, 2025.

‍