AI
AI coding tools and large language models (LLMs) can write code, explain it, and refactor it faster than any person. But writing code was never the hard part of maintenance. Someone still has to understand the system, decide what should change, and live with the result. AI changes how that work gets done. It does not remove the need for it.
Key Statistics
- • 87% of Rails developers say AI tools are either regularly part of how their team works (57%) or something they are actively experimenting with (30%). / 2026 Ruby on Rails Community Survey
- • 71% of Rails teams use AI for code review or test generation, not just for writing new code. / 2026 Ruby on Rails Community Survey
- • 74% of Rails developers say they have adapted how they work and AI genuinely makes them more capable. / 2026 Ruby on Rails Community Survey
- • 50% of Rails developers say the pace of AI advancement makes them anxious about the long-term value of their skills. / 2026 Ruby on Rails Community Survey
- • Experienced open source developers took 19% longer to finish tasks with AI tools, while believing the tools had made them faster. / METR Developer Productivity Study, 2025
Why This Matters
AI makes new code cheap. That shifts the cost of software toward everything that comes after: reviewing it, testing it, understanding it six months later, and deciding whether it should exist at all. Teams that only measure how fast code gets written will miss where the real costs land.
Used well, AI is a strong maintenance tool. It can map an unfamiliar codebase, draft tests for code nobody wants to touch, and carry out large refactors one careful step at a time. Used carelessly, it produces more code than anyone can review and turns a legacy system into a bigger legacy system.
On the Maintainable Software Podcast, language creators, framework authors, and practitioners share how AI is changing their day-to-day work, and which old habits matter more now than ever.
Episodes on AI
EP-230 | October 13, 2026
Gabriela Moreira: When 800 Tests Aren’t Enough
EP-229 | October 6, 2026
Matija Šošić: Abstractions Aren’t Going Anywhere
EP-228 | September 29, 2026
José Valim: Designing a Language You Can Maintain for Decades
EP-227 | September 8, 2026
John Athayde: Maintaining the User Side of Software
EP-224 | August 4, 2026
Diana Pfeil: Building Confidence in Probabilistic Systems
EP-217 | January 21, 2026
Brittany Ellich: Using AI to Maintain Software, Not Rewrite It
EP-190 | November 12, 2024
Justine Gehring: Refactoring Software at Scale with AI
EP-180 | August 20, 2024
Obie Fernandez: Pioneering AI in Ruby on Rails Development
EP-166 | April 16, 2024
Esther Olatunde - AI's Role in Evolving Software Engineering Practices
EP-163 | March 5, 2024
Kyle Daigle - Scaling Up with AI: A New Era of Code Maintenance
Frequently Asked Questions
Does AI-generated code make software harder to maintain?
It can. AI makes it easy to add code faster than a team can review or understand it, and every line added is a line someone has to maintain. The risk is not the tool itself but skipping the review, tests, and design thinking that keep a codebase healthy. Teams that hold AI-written code to the same standards as human-written code avoid most of the problem.
How can AI help with legacy code?
AI is useful for reading code as much as writing it. It can summarize unfamiliar modules, trace how data moves through a system, draft characterization tests before a change, and carry out repetitive refactors across many files. The best results come from small, verifiable steps rather than asking a model to rewrite a system in one pass.
How do you test software that uses an LLM?
LLM-powered features can change behavior without a code change, so traditional pass/fail tests are not enough. Teams use evals: curated sets of inputs with expected qualities, scored automatically and tracked over time. Pair evals with production monitoring, and prefer the simplest model that does the job so there is less to go wrong.
Do frameworks and abstractions still matter if AI writes the code?
Several guests argue they matter more. Frameworks give both people and AI agents a shared set of conventions, which makes generated code more predictable and easier to review. Good documentation and clear error messages, long valued by developers, turn out to be just as valuable to coding agents.
Will AI replace software developers?
Our guests do not expect it to. AI handles more of the typing, but someone still has to understand what the software should do, judge whether a change is correct, and own the system over time. Those skills, especially understanding how software should behave, are becoming more valuable, not less.