🤬
04.03.2026 13:45
👍 0
🔁 0
💬 0
📌 0
🤬
"Make people dependent enough and then? Make it shitty!"
A Day in the Life of an Ensh*ttificator
youtu.be/T4Upf_B9RLQ
« Three architectural principles are gaining consensus: "token efficiency" as a design constraint (code structure), explicit over implicit everywhere (types, error handling), and co-location of related code. [they] aren't new, but AI has given them renewed urgency. »
www.cjroth.com/blog/2026-02...
I spent the last days writing a lengthy blog post about my AI journey on a brownfield modernization project, focusing on the last six months of it where I heavily used Claude Code, and how my workflow changed completely. Enjoy!
www.yduman.dev/posts/six-mo...
Disagree. ML is a subdiscipline of AI.
What are your tools and approaches you're working on?
I've been working in the intersection of AI, Data Science and Legacy Modernization for over a decade now. Now it's time to take it even further with LLMs to make AI-assisted Software Evolution work to tackle legacy software modernization (and have fun playing around with new toys at the same time ☺️)
Yes, models evolved great in recent times and the data from the study is unfortunately very dated (Q1 2025 when I'm not mistaken). That means pre Claude Code / Agentic AI era. I hope someone will provide some fresh data in the near future.
Now updated with new colors and without misleading numbers.
I don't build AI into legacy systems. I use AI engineering for building analysis and modernization tools on greenfield using Python/pandas/Neo4j/OpenRewrite for analyzing legacy systems.
Why does working with AI agents in legacy systems feel so much harder than in all those vibe coding demos on YouTube? 🤔
Part of the answer:
😊 Popular language + Greenfield ☘️ = "AI heaven"
😯 Niche language + Brownfield 🟫 = "AI hell"
More about this topic: markusharrer.de/blog/2026/02...
AI assistants that write emails, shop online, or book travel may feel convenient — but they’re often insecure by design. In her blog post, Joy Heron offers an important reality check on the current hype.
📖 Read now: www.innoq.com/en/blog/2026...
Wake me up once your AI agents are actually fixing legacy systems, not just analyzing them.
The picture looks different for 754 Java files, 8711 methods, and 82830 SLOCs. But there are still method clusters (that don't fit the package structure at first glance).
Here I demo how you can use machine learning techniques to visualize your legacy code on another abstraction layers with "method clusters":
Left: structural similarity
Right: functional / business purpose (via LLM)
Colors: The real domain-oriented modules
More: github.com/feststelltas...
Here is my rough prototype that parses legacy RPG/SQL source code, extracts dependencies, and loads them into Neo4j. It then runs clustering algorithms to identify isolated code segments (incl. LLM summarization) that can (hopefully) be migrated to another language.
github.com/feststelltas...
Probably the two most common Bing searches in Microsoft Edge are:
1. "Firefox"
2. "How to change the default browser in Outlook"
Dependency cruiser is nice. An elegant way to express and enforce architectural constraints. github.com/sverweij/dep...
It's not just "Human 🧒 vs. AI 🤖" these days when you're thinking about transforming legacy code on a larger scale.
That is why I visualized a part of the refactoring tools landscape using my favorite thinking tool: PowerPoint 🧡
That was exactly the feeling I had after hearing this from a Claude Code developer 😅
www.youtube.com/watch?v=8qtS...
Don't worry. Claude Code uses "agentic search" (which is using, well, glob and grep under the hood).
Why We've Tried to Replace #Developers Every Decade Since 1969
https://www.caimito.net/...
Stop Hoping Your Architecture Holds: Enforcing Quarkus Design Rules with JQAssistant
How to turn architecture guidelines into executable rules using Quarkus, JQAssistant, and graph-based analysis
www.the-main-thread.com/p/quarkus-jq...
Only if you maintain it!
Otherwise, it’s just a data transform.
Data, of course, is the real liability.
Hey, my talk "6 Safe Refactorings for Untested Legacy Code" that I gave at @confooca.bsky.social 2025 is available online!
How cool is that? 😄
So here it is: I'm presenting you with 6 moves you can use in VS Code to refactor untested code ✊
www.youtube.com/watch?v=5YEm...
We have started preparing our 2026 edition 🤫
In the meantime, you can find all the talks from previous editions on our YouTube channel: www.youtube.com/@mendercon/v...
Stay tuned to learn more about what's coming 😉
#MenderCon #LegacyCode
And it's up! All together now...
🎶 Somebody told me
the user provider
should use an adaptor
to proxy the query
factory builder... 🎶
www.youtube.com/watch?v=p03o...
Certainly I am not the only one to notice, but "ai" is right there in "naive."
"Starting now, you’ll be able to make a career out of un-LLM-ifying applications." Psst 🤫, don't tell everyone!
www.oreilly.com/radar/job-fo...