mg

mg

@morng.bsky.social

Developer of 18XXc & BankBuilder

84 followers183 following272 posts

LOL, Is this some LLM slop, or is this actually how Easter Sunday is calculated for the year? What did people do in the medieval times?, just guess?

2026 has been emotionally exhausting. When pondering my future as a software developer I am both these guys, switching seats sometimes multiple times per day.

LLM generated Javascript vs LLM generated Elixir in the same file from the same prompt and model. One looks fine, one looks like the slopiest slop south of Sloptown. #ElixirLang

LLM's are the best code commenters in the world, if you take the time to specify clearly what you want and dont want in a comment. If you make the LLM write the code, make sure you make it write good comments as well.

The throw - catch and raise - rescue dichotomy in #ElixirLang is very confusing. Am I supposed to just know if some library is throwing or raising errors?

Newest version of 18XXc as a new feature. It is now possible to directly input the route value instead of composing it from the stops and bonuses. It takes advantage of the fact that route totals are always a multiple of 10, so it automatically multiplies the input by 10. #18xx

It is easy to switch between the two methods, using any of the stops/bonuses buttons switches to the old way of calculating, and pressing the total amount switches to the new method of calculating.

It came about because a user asked if it was possible to directly input the totals. I had been thinking about this for some time, and have further ideas on how to make this method more powerful in the future.

Hi @zachdaniel.dev would you mind sharing what technology you are building your AI stack in? Im guessing elixir and ash, but are there some other key pieces that you would be willing to share? Need to build something similar, would appreciate the info.

SmartLogic@smartlogic.io

🤖 🧙‍♀️ Today, Zach Daniel joins Elixir Wizards to walk us through LLM integration at an organizational level and why teams shouldn't "sell their organization's spinal column" by outsourcing the context layer around their AI tools. 🎧️ bit.ly/EWS15E12 📺️ youtu.be/Is1Dg6ywC8U #elixirlang #myelixirstatus

18XXc v3.19 is out, adds support for 1822CA, 1822PNW, 1832, 1835, 1880, 1888-N, 18Cuba, 18DO, 18EU, 18Espania, 18FL, 18GA, 18GB, 18IN, 18NY, 18RoyalGorge, 18USA, 21Moon, Harzbahn 1873, Railways of the Lost Atlas. #18xx

What is a good software package for creating slides from Markdown files? Maybe even for presenting inside a TUI.

When building a GenServer, some handle_call's belong with some handle_info's. I wish the compiler could handle specifying them together, instead of forcing me to group all handle_call's together, all handle_info's together. To me this is sacrificing code structure to benefit the compiler #ElixirLang

Coding agents have solved the 80/20 Pareto Principle in coding. Now we live in a 99/1 Pareto Principle world.

Model your domain, ask for the rest. I think 95% of CRUD busywork might just go poof with the combined powers of #Ash and LLMs. This is unreasonably effective. #ElixirLang

Are Steve Yegge and Geoffrey Huntley the prophets of this new mad age? Or have they gone totally of their rockers? Or both?

What Im thinking now about LLM coding is; sure, it is liberating not being held back by gaps in knowledge; yes I can now easily build 10k LOC in 1 week; but can I bring it over the line? Can I finish the work and create something useful? Or will it crumble under the weight of accumulative complexity

My Claude Code is using phrases like "my instict is.." and "my preference is..." Is this thing just going to will itself into consciousness?

I know its not very important, but I wish I could do descibe "..." do describe "..." do .. end end #ElixirLang

Claude Code is an excellent teacher. I thought I had a problem with my Ash validations (argument not allowed to be null, but the resource was created when missing) but turned out to be a known and defined behaviour with Ash.Geneartors The more you know, and the sooner you know it. #ElixirLang

note I didnt tell Claude what the problem was, just that there might be one. It figured out the rest on its own.

Claude Code generates #ElixirLang and Ash code like a champ, and it uses Tidewave MCP search the docs and to test all validations and changes and whatever else I ask it to build. I feel Ash/Elixir are designed to encourage building bottom up, piece by piece, and Claude is very efficient for that.

TreeSitter and LSP functionality is not very good with Phoenix CoLocated hooks. Is there some way to use src to refer to a js or ts file that is colocated with the LiveView heex file? #ElixirLang

Using a 'zipper' just to traverse a list in both directions is so unnatural to me as someone that is used to random access data structures. My brain is still not wired for this. #ElixirLang

Played a game of 18SJ. The incremental phase slows the game down IMO but we could probably push the trains faster. 3 companies got nationalized, 1 by design and 2 by accident. The yellow is helpful but maybe it guides the map to a similar shape every game? #18xx

This came to my head as I was thinking on how to use agents to help me build a Ash/Phoenix/Elixir side project. Given how data oriented these stacks are, I think it might be not that far from the truth.

I've watched @mitsuhiko.at latest video on his way of doing agentic coding. Watched it 3 times already. Today I tried it. I wrote a spec for a feature. I asked Claude to read it and tell me what it thought. Had a back and forth, Claude would come in with questions, I would update the spec to...

View full thread

Most of the value here is talking to claude about the feature, it asked a lot of questions that I had not considered and allowed me to build a more robust version than I would have otherwise.

I've watched @mitsuhiko.at latest video on his way of doing agentic coding. Watched it 3 times already. Today I tried it. I wrote a spec for a feature. I asked Claude to read it and tell me what it thought. Had a back and forth, Claude would come in with questions, I would update the spec to...

...clarify, tell claude to reread, would get more questions, back and forth. After a few iterations I felt I had clarified what I needed, and asked claude to build a plan for building the feature according to the document. Claude did, I modified it a bit, and then I asked Claude to tell a...

subagent to build the feature. The feature got built, and works as I wanted. The spec is 80 lines of text, the plan about the same, the code about 120 lines of #elixir code, and 51 tests. It took about 2 hours, and since I dont really know Elixir I am sure it would have taken longer to do it alone

Is the Apple Watch automatically deleting podcast episodes (that I have not listened to), or is it an Overcast issue? I cannot believe how bad the process of downloading episodes is on AW, and this "cleanup" is making everything worse. This thing has WiFi and 5G, why is is so bad at downloading

Just finished season 2 of The Leftovers. I am still very confused with where this show is going; I don't understand the point of many plot lines and the motivations of most characters remains a mystery to me. But the show is very good at evoking sadness. Onwards I guess.

I used Claude Code to search through my entire @obsidian.md for certain phrases (and include common variants and misspellings) in both path and node contents. Then I told it to send the entire list of matched files to a running neovim instance in another tmux window as an arglist. Amazed.