For those of you who are developers, either professionally or as a student, hobbyist, or mere dabbler... What AI coding tools do you use (or have you tried)? What type of AI tool is it (IDE extension, custom IDE, CLI agent, web-based app builder, etc)? Which backend LLMs did you use or can you use within that tool (Claude Sonnet, Gemini Pro, OpenAI Codex, Grok Code, etc)? What specific version of the model did you use, if you know (i.e., Claude Opus 4.1, Claude Sonnet 4, Claude 3.7 Sonnet, etc)? What is the price/cost for the AI setup you use (subscription costs of the tool and/or LLM fees)? Is there a free tier, and if so, did it allow you to do very much... or did you quickly have to migrate to a paid subscription to do anything "real"? How difficult was it to set up and use? Did you have to customize? Do you use multiple tools (i.e., Claude code within Cursor)? How good were your results (i.e, "It was amazing and one-shot my app based on a single prompt!", "It was easily distracted and 'hallucinated' weird solutions.", "Its code was nearly useless and/or poor quality.", "It wasn't responsive to my prompt at all, and it built something I didn't want or ask for.", "It built outdated solutions using deprecated code/frameworks.") What was your "use case" (i.e., web front-end development, local GUI app for Windows/macOS, local CLI app, etc)? What language or frameworks were used? It's the "wild west" in generative AI code development right now, and tools/practices are rapidly changing. There are so many companies and tools being rushed to market that it's hard to keep track of them all. I'm just curious as to what people's experiences have been and what they have tried. Please note that some of the listed tools may be slightly misclassified or have been changed/deprecated, as I have not used all of them and am not omniscient about them.
Can't help you with this, since I don't code. But what is your opinion, do you think this is the future? Many big companies like Microsoft and Salesforce claim that a lot of coding is already done by AI, so they have fired thousands of software developers.
My answer is no. Not a developer, but if and when I need to write a snippet of code, I want to do it the hard way, to fully learn and understand what gives. In other words, I will never deploy something I don't understand how it works. It's also a paradoxical thing: AI can help people "quickly hack" solutions, but there are no shortcuts in life. If anything, it's the seasoned people who could/should use AI tools, except these tools are not good enough to really help the seasoned people, as they need to double-check the AI work. Technically, it's called AI, but it's no different than macros, reusable code and such. If you're already well versed in the craft, you already have all these tools and automation as part of your workflow. Thus, in my view, is the problem of wrong tool for the wrong people, right tool for the wrong people, or wrong tool for the right people. Perhaps in 10-15 years, AI will be useful, but as long as false positive rate > true positive rate in terms of code quality and error by at least an order of magnitude worse than what humans produce, there's no real value to it. If an AI tool is "only" say 30% better than the human, it means you need to check the output 7/10 times, or correct 7/10 times. In other words, the error rate is so big, you don't really win anything as you need to double-check the code. It only makes sense once the error becomes so small you can truly "relax" with the output. Otherwise, you might as well make all code with AI and have humans spend all their time fixing and checking it. Mrk
