“When you spend years responding to problems, you can sometimes overlook the fact that you could be preventing them.”
— Dan Heath

‍

In Issue 012, we talked about the inconvenient truth that most modernization efforts fail in part because they’ve lost sight of the “purpose of software, which was to serve as a backbone to the organization, in all its organic and chaotic human-factor glory, to help it work smarter, faster, and more flexibly.”

‍

Quality and malleability of software have always mattered, but not necessarily enough to break free of immediate demands, whether it’s triage modernization, or shortcuts made in the face of political expediency. These incentives form a kind of inertial gravity, an organizational molasses that can only be disrupted by, say, an incoming meteor named “AI”.

‍

This same, fiery sea change exposes a tantalizing possibility: what if AI could solve our legacy modernization problems?

‍

Reader, behold the second inconvenient truth: AI will not solve your legacy modernization problems.

‍

For one, AI doesn’t solve the incentive issue: modernization remains a high-risk, low-reward endeavor that fazes all but the most intrepid technology leaders. Even if we could trust AI-generated code, which is inherently non-deterministic and hallucinates today, it still must exist within a methodology designed to reduce risk and disruption, especially for the most critical systems. (Obviously, many teams, including us, are working on this: in January, Codescene wrote about refactoring JavaScript.)

‍

More ontologically, legacy modernization isn’t something to be solved. There’s no end state to modernization, there is only the ability to minimize the cost of change in the future.

‍

No one sets out to make short-shelf-life software; rather, we’re always working with a set of constraints, incomplete information, conflicting mandates. Whose product is prioritized to get to market faster? Which board member is barking the loudest about a particular risk?

‍

And that is why there is no modernization finish line. The goal isn’t a milestone, it’s a state of being: a codebase that’s easy to understand and easy to change, that’s responsive to business needs, competitors, regulatory zigs and zags.

‍

So consider every limping-but-not-entirely-broken application, every looming compliance or upgrade deadline, every reputation-wrecking IT outage, as opportunities to evolve into a state of continuous, minimally stressful, perhaps even cautiously joyful, modernization.

‍

News and Views

Accumulated tech debt leads to unhappy engineers. Unhappy engineers lead to application failures. Application failures lead to plummeting share prices. Sonos learned this the hard way.

‍

Not entirely a surprise: AI doesn’t help with tech debt, since the drivers of tech debt are not solely technical.

‍

Think small. The latest DORA report notes, “Contrary to our expectations, our findings indicate that AI adoption is negatively impacting software delivery performance.” They hypothesize that the enthusiasm for AI “may have caused the field to forget one of DORA’s most basic principles—the importance of small batch sizes.”

‍

From the Orchard

We spoke with Margaret Exley, CBE, about the view from the board of technology risks in our latest On a Limb podcast. Delightful phrase of the month: “Someone has to ask the idiot questions.”

‍

The fabulous Dave and Dharm hosted me on the latest episode of their Dave and Dharm DeMystify podcast, where we riffed about socio-technical system design and why code-first approaches to modernization don’t work very well.

‍

We’re at Evanta's Houston CIO Executive Summit on December 10th. Edward Hieatt will be there to talk about AI, modernization risk equations, and building confidence through tangible progress. Attendance is by invitation only: apply here or send a direct email to the Evanta community programs manager.

‍

Going to AWS re:Invent December 2-6th in Las Vegas? We’ll be there too. Let’s meet up.

‍

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

‍

——————

Issue first published on November 26th, 2024.

‍