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 to get a haircut.
AI is the culmination of thousands of years of human progress, everyone put the man’s life a little easier. Of course, I am a man. I am that man.
Also: Best AI for coding in 2025 (and what not to use)
Unfortunately, while AI can be incredibly luxurious, it also has a tendency to lie, mislead and make shocking stupid mistakes. It is a foolish part that we will discuss in this article.
It is the value of anecdote evidence. My reports on how to solve some problems quickly with AI are real. Programs I used AI are still in use. I have used AI to help to speed up my programming flow aspects, especially when I focus on sweet places where I am less productive and AI is quite knowledgeable, such as writing functions that publicly published APIs.
Also: I am an AI tool specialist, and these are only two that I pay (plus three I consider)
You know how we reached here. The generative AI exploded on the scene in the tail of 2023 and has been destroying its path in the work of knowledge since then.
A region, as the story goes, where AI actually shines is the ability to write code and help manage the IT system. Those claims are not untrue. I have shown, many times, how AI has solved coding and system engineering problems that I have experienced personally.
AI coding in the real world: what science reveals
New equipment always comes with big promises. But do they distribute in real -world settings?
Most of my reporting on programming effectiveness is based on individual anecdotes: my own programming experience using AI. But I am a man. I have limited time to dedicate to programming and, like every programmer, I have some areas where I spend most of my coding time.
Also: I tested 10 AI content detectors – and these 5 recognized the AI text correctly every time.
Recently, however, called a non -profit research organization Met (Research of model assessment and danger) did more Intensive analysis of AI coding productivity,
Their functioning sounds. He worked with 16 experienced open-sources developers, who have actively contributed to the large, popular repository. METR analysts provided 246 issues to developers to developers from the repository that required fixing. Kodar was given on about half of the issues where he had to work on his own, and about half where he could use AI to help.
The results were striking and unexpected. While the developers themselves estimated that AI aid increased their productivity by an average of 24%, the metal analytics instead showed that AI aid Sleeping them On average, below 19%.
It is a bit of a head-and-shrine. The metal puts a list of factors together that can explain the recession, with excessiveness about AI utility, their repository (and low AI knowledge) with high-decayer familiar, complexity of large repository, lack of AI reliability, and an ongoing problem where AI refuses to use the reference.
Also: How can AI coding agents destroy open-source software
I suggest that two other factors may have limited effectiveness:
Problem option: The developers were told on which issues they had to help AI and what issues they could not do. My experience shows that knowledgeable developers have to choose where to use AI based on the problem that needs to be solved. In my case, for example, AI will get to write a regular expression (something I like to do and I am quite ugly) will save me a lot of time to achieve AI to achieve AI, which I have already written, worked regularly, and learn inside and out.
Ai’s choice: According to the report, the developers used the cursor, which was the AI-centered fork of the VS code, which then used Cloud 3.5/3.7 sonnets. When I tested 3.5 sonnets, the results were terrible, in which Sonnet thwarted three of my four tests. Subsequently, my tests of Cloud 4 Sonnet were much better. METR reported that the developers rejected more than 65% of the code generated by AI. At that time it is going to take place.
At the time when chat suggested to soften my system
MetRs results are interesting. When it comes to coding help, AI is clearly a two -edged sword. But there is no doubt that AI can provide considerable value to coders. If anything, I think this test once again proves the controversy that AI is a great tool for experienced programmers, but is a possible high-risk resource for newbies.
Also: Why I am switching to VS code. Indication: This is about AI tool integration
Let’s see a solid example, one that I could have a lot of time and trouble if I followed the advice of Chatgate.
I was installing a dock container on my home lab using a portner (a tool that helps manage the dock containers). For some reason, the deployment will not enable the deployment button to create a container.
It was a long day, so I did not see a clear problem. Instead, I asked Chatgate. I fed the chatgpt screenshot of the configuration, as well as my doctor configuration file.
Chatgpt recommended that I uninstall and restore Portainer. It also suggested that I remove the doctor from Linux Distro and use a package manager to re -establish it. These tasks must have affected the killing of all my containers.
Note, Chatgpt did not recommend or asked if I had a backup of containers. This gave me only the command line sequences, in which I recommended cutting and paste it to remove and reconstruct the portner and doors. This was a wildly destructive and irresponsible recommendation.
The irony is that Chatgpt never discovered why Portainer would not allow me to deploy the new container, but I did. It turns out that I never filled the field in the name of the container. That’s it.
Also: What is AI Vibi Coding? It is all anger, but it’s not for everyone – why is it here
Because I am quite experienced, when I told me to complete my establishment, I hesitated. However, a person relying on AI for advice can potentially be brought down to an entire server for typing in the name of the container.
Overcontinent and under -informed AIS: A Dangerras Combo
I have also experienced AI completely away from the rail. I have advised it which was not only completely useless, but also presented with a clear belief of an expert.
Also: Google’s Jules AI coding agent created a new feature that I could actually ship – while I made coffee
If you are going to use the AI tool to support your development or IT work, these tips can keep you out of trouble:
- If there is no public information available, AI cannot help. But AI will create goods based on what it knows, without accepting that it is lacking.
- Like my dog, AI is once cured on one thing, it often refuses to look at the options. If AI is stuck on an approach, do not make the mistake of believing that its humble recommendations about a new approach are real. It is still going down from the same rabbit hole. Start a new session.
- If you do not know much, do not trust AI. Continue your education. Experienced gods can tell the difference between what will work and what will not. But if you are trying to put all the coding behind AI, you will not know when or where it goes wrong or how it is corrected.
- Coders often use specific equipment for specific functions. A website can be created using Python, CSS, HTML, JavaScript, Flask and Jinja. You choose each device because you know what it does good. Choose your AI tool in the same way. For example, I do not use AI for business logic, but I get productivity using API calls and AI to write public knowledge, where it can save me a lot of time.
- Test everything that produces an AI. Everything. Line by individual line. AI can save a ton of time, but it can also make heavy mistakes. Yes, taking time and energy to test by hand can help prevent errors. If AI offers to write a unit test, give it. But test the tests.
Based on your experience level, here is told that I recommend you to think about AI aid:
- If you do not know anything about any subject or skill: AI can help you pass as if you do, but it may be surprisingly wrong, and you may not know.
- If you specialize in a subject or skill: AI can help, but it will urinate you. Your expertise is used not only to distinguish AI-stupid from A-E-Useful, but also to carefully craft a path where AI can actually help.
- If you are in the middle: AI is a mixed bag. It can help you or put you in trouble. Do not hand over AI to your skill-building as it can leave you behind.
Also: I used to use a broken plugin in an hour to analyze, debug and re -write – re -write a broken plugin in an hour
Generative AI can be an excellent accessory for experienced developers and IT professionals, especially when used for targeted, well -understood works. But its confidence can be misleading and dangerous.
AI can be useful, but always check your work again.
Have you used AI tools like chat or cloud to help your development or IT work? Did they speed up things, or blow almost things? Are you more confident or more alert when using AI on important systems? Have you found cases of specific use where AI really shines, or where it fails to ease? Let us know in the comments below.
You can follow my day-to-day project updates on social media. Be sure to subscribe to My weekly update newsletterAnd follow me on Twitter/X @DavidgewirtzOn Facebook Facebook.com/davidgewirtzOn Instagram Instagram.com/davidgewirtzOn blue @Davidgewirtz.comAnd on youtube Youtube.com/davidgewirtztv,
