This is what a lot of code will look like in a couple of years.
07.03.2026 09:47
👍 0
🔁 0
💬 0
📌 0
This is what a lot of code will look like in a couple of years.
And again blindly relying on Claude can get you in serious trouble. This was plan wrong!!
Amplifying learning, I am all with you. For that purpose AI can be very effective. But again quite dangerous as well. Beware of the brain rot😜. There is also literature how AI can diminish critical thinking and intellectual skills.
That is so true. And something that also impacts usability of AI. I haven't seen cases where the AI said, I have added this feature and also abstracted some previous code away. The opposite seems to happen.
That is a very valid concern! That's why I created a workflow that prevents repo-assist to directly meddle with my source code. Instead I have it make coding suggestions implemented in isolated scripts. But besides that, with respect to issues handling, documentation etc.. repo-assist is a great!
I think this is one of the most valuable #fsharp libraries. Would it help to see if the repo-assist from @dsyme.bsky.social could help? With some explicit rail-guards it is very useful for my project: github.com/informedica/.... Preventing direct edits of my source code.
Hard problems require hard work. No LLM is going to fix that. I have burnt myself a couple of times to learn that lesson.
I am getting very enthusiastic👏 about this. Repo-assist has now labeled all my issues as a first step.
Now, extrapolate that to all the claims that AI is going to take over software development and all sorts of other intellectual jobs. ???
So, just adding a column to a markdown table turns out to be a massive challenge not only for Claude Opus 4.6 but also GPT-5.2 is really struggling. Note that it perfectly "knows" what to do, only the actual adding of the column to a otherwise simple table seems a real problem.
Everybody is talking about how AI is going to take over programmer jobs. I just asked Claude Opus 4.6 to align markdown tables in a document. A very simple and humble question. It proceeded with some python script that really messed up things. I think I will continue writing my own code.😅
The reason: "If you were primarily working with C#: Use our newly introduced file-based apps!" This doesn't bode well for any other language than C#.
I can understand that, but doing coding myself (at least for me) is essential in getting to the right solution direction. Mozart could compose a piece right from his head, I am no Mozart ;-).
As for real guards, a strongly typed domain is the best you can get and that is a strong point of fsharp.
I have the luxury to be able to choose my language. And I hope that I can afford more F# developers in the near future! Depends on my institution whether they will get the financing.
It could also be the next million dollar mistake. I have had mixed results with AI coding. It's fine with established practices, but if you want to create something new, that's a whole other thing.
I don't use Nuget, I use paket. I like the direct manual management better using dependency files.
I do use Rider as well. But that's because I happen to have a full subscription and I have need of different language IDEA's. VS Code with Ionide is my prefered #fsharp code editing IDEA though.
I manage this repository using the Ionide/VS code: github.com/informedica/.... Granted this could be a case of "this runs on my computer so what are you complaining about". Also, I do a lot of building testing command line. Code generation always using scripts except for the UI.
I hope not as I have based a large project entirely on #fsharp: github.com/informedica/...
It would be strange to condemn #fsharp to death while every other language tries to emulate functional programming concepts. As to IDEA support, last couple of years VSCode is really stable now. Rider does do a good job (aside from some annoyances). Also, I successfully can run F# clientside.
And again Rider keeps adding unsolicited opening statements creating naming conflicts. This is driving me crazy as I have reported this again, and again and again and again....
This could be the next null value mistake!
I thought it was just me really disliking the code generated by Claude. Also, a realization that when coding I often reflect on existing code and always find ways to improve that code, something any AI seems entirely incapable of.
I told claude code to NEVER EVER touch my #fsharp code base files, i.e. anything *.fs, but instead do all implementations in script files, so I can first check what is going on. And I am so glad I did this.
Hello @ursenzler.bsky.social. The troubles are that the way you can customize Rider is very confusing. For example, disabling code completion using AI. Trying to find that is really difficult. And a lot of these "useful" functions seem to be turned on by default an difficult to turn off.
Hello @auduchinok.bsky.social . The problems I have are related to use of code completion and intellisense. In general Rider works fine, but somehow, there are a lot of "magical" things happening that really get into the way. There are just too many settings to get a grip on that.
And again when using Rider with #fsharp, I run into serious usability problems. Right now the only reliable alternative (on non Windows systems) is VSCode. I really hope VSCode, i.e. the Ionide extension keeps working, otherwise I am in trouble.
One thing is sure, when coding #fsharp, Copilot can destroy more code than I care for. So, no more Copilot writing code for me, it's easy enough to write my own #fsharp code. Just using Copilot to write comments, tests or checking coding logic.
My addition for the #fsharpadvent: informedica.nl?p=437. A short blog demonstrating the powers of #fsharp fsi in real life production coding. Thanks @sergeytihon.com !
Eh, yes!! It allows me to write #fsharp everywhere I want.