glitch.com brings back memories of the pre-slack era Glitch MMO. What an amazing time that was, really beautiful game made by talented people, pivoted the communication side into what we now know as Slack
I did :) I made a browser-based MMO with Phoenix to test out liveview and learn the language: https://shopkeep.gg
And it was pretty annoying. Elixir doesn't really lend itself to vibe coding due to namespacing and aliasing of modules, pattern matching, all without static typing (I know, Dialyzer...). It also struggles to understand the difference between LiveComponents and LiveViews, where to send/handle messages between layers.
Without references to filenames, the agent perpetually decides "this doesn't exist, so I'll write it :)". I found it to be pretty challenging before figuring out I could force the agent to run `mix xref callers <Some.Module>` when trying to do cross-module refs.
Apotris (https://akouzoukos.com/apotris), a GBA "demake" of tetris is one of the cooler ones I've seen with regard to QOL. Lots of settings for tweaking controls and sensitivity that make it possible to fine tune the game to just feel fantastic to play
this is a huge issue for me as well. It just kind of obfuscates errors and masks the original intent, rather than diagnosing and fixing the issue. 3.5 seemed to be more clear about what it's doing and when things broke at least it didn't seem to be trying to hide anything.
Cursor has been a really interesting journey for me. I've built an entire Elixir/Phoenix browser MMO, but I have almost no confidence that I've done things idiomatically. Adding rules to urge the agent to be more "idiomatic" doesn't provide confidence.
Additionally, Elixir does not really lend itself to automated coding since the message/event based flows don't make dependencies obvious, and typing is soft and not-enforced statically. Adding rules like "always call mix xref callers <module_name>" or "run compile command" after each change has been helpful, but I still end up with bugs where changes in one part of the app cause runtime errors that are not only tricky to diagnose, but require multiple iterations of logging and manual debugging - which becomes costly when each Cursor agent costs me 4 cents. Perhaps this is standard for Elixir, or maybe I need to be architecting my code patterns in a smarter way.
The most frustrating part is how much of a yes-man the agent is, it would be really nice to be able to have the agent steer me away from decisions where the code is being written against best practices. I agree with the author's sentiment - The "learning moments" I run into seem to be whenever Cursor screws something up and I have to investigate myself, rather than a sort of learning-with-guidance while doing it right.
if you want to see the true horror, check out https://comma.ai. $2000 and plugs into most cars made in the past few years, works by using cracked security for the cars it is "compatible" with. These people are on the road next to you with a car being driven by a single smartphone camera. They sell it as "chill driving" but they have a discord where people just flash custom FSD firmware.