Our story begins, as many stories do, with a man and his AI. The man, like many men, is a bit of a geek and a bit of a programmer. He also needs a haircut.
The AI is the culmination of thousands of years of human advancement, all put to the service of making the man’s life a little easier. The man, of course, is me. I’m that guy.
Also: The best AI for coding in 2025 (and what not to use)
Unfortunately, while AI can be incredibly brilliant, it also has a propensity to lie, mislead, and make shockingly stupid mistakes. It is the stupid part that we will be discussing in this article.
Anecdotal evidence does have value. My reports on how I’ve solved some problems quickly with AI are real. The programs I used AI to write with are still in use. I have used AI to help speed up aspects of my programming flow, especially when I focus on the sweet spots where I’m less productive and the AI is quite knowledgeable, like writing functions that call publicly published APIs.
Also: I’m an AI tools expert, and these are the only two I pay for (plus three I’m considering)
You know how we got here. Generative AI burst onto the scene at the cusp of 2023 and has been blasting its way into knowledge work ever since.
One area, as the narrative goes, where AI truly shines is its ability to write code and help manage IT systems. Those claims are not untrue. I have shown, several times, how AI has solved coding and systems engineering problems I have personally experienced.
AI coding in the real world: What science reveals
New tools always come with big promises. But do they deliver in real-world settings?
Most of my reporting on programming effectiveness has been based on personal anecdotal evidence: my own programming experiences using AI. But I’m one guy. I have limited time to devote to programming and, like every programmer, I have certain areas where I spend most of my coding time.
Also: I tested 10 AI content detectors – and these 5 correctly identified AI text every time
Recently, though, a nonprofit research organization called Meter did a more thorough evaluation of the threat. A thorough analysis of AI coding efficiency
Their method seems sound. They worked with 16 experienced developers who actively contributed to large repositories. METR analysts gave these developers 246 issues to fix from repositories. About half of the issues were for coders to fix on their own and half could be solved with AI assistance.
Results were surprising and shocking. While the developers estimated that AI assisted them by an average 24%, METR’s analytics showed that AI assistance actually slowed them down by an avg. 19%.
This is a bit of a puzzler. METR compiled a list of possible reasons for the slowdown. These include over-optimism regarding AI’s usefulness, a high level of developer familiarity with their repositories and less AI knowledge, the complexity of large repositories and faulty AI. “important tacit knowledge or context.”
See also: How AI coding agent could destroy open-source code
Two other factors may have limited effectiveness. My experience suggests that knowledgeable developers should choose where to use AI depending on the problem to be solved. In my case, having the AI write a regular-expression (something I hate to do and am not very good at) would save me more time than having the AI modify code that I have already written, worked on regularly, and understand inside out.
Choose AI: The report states that the developers used Cursor which is an AI-centric fork from VS Code. At the time, Claude 3.5/3.7 Sonnet was used. When I tested 3.5 Sonnet the results were horrible, with Sonnet failing to pass three of four of my test. My tests of Claude Sonnet 4 were much better. METR reported developers rejected over 65% of code generated by the AI. It’s going take time.
The time ChatGPT suggested nuking the system.
The METRs results were interesting. AI can be a double-edged weapon when it comes coding assistance. There’s no doubt that AI is a valuable tool for coders. This test, in my opinion, proves that AI is not only a valuable tool for experienced programmers but also a high-risk resource for beginners.
I’m also switching to VS Code. Hint: It is all about AI tool Integration
Let me give you a concrete example. I could have saved myself a lot of trouble and time if I had followed ChatGPT’s advice.
Portainer, a tool for managing Docker containers, was helping me set up a Docker Container on my home lab. Portainer refused to enable the Deploy button in order to create the container.
I had a long day and didn’t see any obvious problems. I asked ChatGPT instead. I sent ChatGPT screenshots and my Docker configuration as well as my Docker file. ChatGPT suggested that I uninstall Portainer and reinstall it. It also suggested that I remove Docker and reinstall it using the package manager. These actions would have killed all my containers.
ChatGPT did not ask or recommend that I have backups of my containers. It only gave me the command-line sequences that it recommended I copy and paste to delete Portainer and Docker. It was a destructive and irresponsible suggestion.
It’s ironic that ChatGPT didn’t figure out why Portainer wouldn’t let me deploy the container, but I did. It turns out that I never filled in the container’s field. That’s it.
What is AI Vibration Coding? It’s popular but not for everyone.
I hesitated because I am fairly experienced when ChatGPT said to nuke my install. Someone who relied on the AI’s advice could have brought down an entire system for not typing in a container name.
Overconfident AIs and Underinformed AIs are a dangerous combination
Similarly, I’ve seen the AI go completely off the rails. I’ve seen it give advice that was not just completely useless, but presented with the confidence of an expert.
Google’s Jules AI coder built a feature I could ship while I brewed coffee.
These tips can help you avoid trouble if you use AI tools for your IT or development work. The AI will invent things based on the little it knows without admitting it lacks experience.
Depending on your experience, here’s what I recommend:
- if you don’t know anything about a skill or subject: AI could help you pass the test as if it did, but you might not realize.
- AI can be helpful, but it may irritate you if you are an expert on a particular subject or skill. Your expertise is used to not only separate the AI-stupid and the AI useful, but also to carefully craft a way where AI can help.
- You’re somewhere in the middle: Artificial intelligence is a mixed bag. It could either help you or get into trouble. You don’t want to leave your skill-building up to AI. It could leave you behind.
How I used ChatGPT in an hour to analyze, debug and rewrite a plugin from scratch.
Generic AI can be a great help for experienced developers and IT professionals, especially when used on well-understood, targeted tasks. Its confidence can be dangerous and deceptive.
AI is useful, but you should always double-check the work it produces.
Do you use AI tools such as ChatGPT or Claude for your IT or development work? Did they help you or did they make things worse? Are you more or less confident when implementing AI in critical systems? Have you come across specific cases where AI excels or fails hilariously? Comment below to let us know. You can follow me on social media for my daily project updates. Subscribe to Follow me on Twitter/X and subscribe to my weekly update newsletter (). @DavidGewirtzon Facebook at Facebook.com/DavidGewirtzon Instagram at Instagram.com/DavidGewirtzon Bluesky at @DavidGewirtz.com (19459058) and on YouTube: YouTube.com/DavidGewirtzTV.
