Be advised: this thread is not for serious in-depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.
- 111
- 2
Jump in the discussion.
No email address required.
Notes -
An update on my Rimworld modding-spree:
Holy shit. Opus 5.5 is just that good. I'd heard people gushing about it since release, and I've definitely tracked the flood of new software it and Astra enabled. I just hadn't used it for anything more intensive than what Fable or prior Opus models had been capable of.
I got my hands dirty by starting a mod that expands the permit system from the Royalty DLC. It worked so smoothly I was taken aback, and my involvement was largely playtesting, ideation and balance suggestions. Even the testing was largely automated, Claude could write test scenarios, take screenshots and review logs.
We went on to MVP - by which I mean the maximally viable product. It had no issues integrating content from 45 additional mods when I shared my 500+ mod list.
It also made permits much more flexible, and added customization and QOL no problemo.
I had been tracking token usage, and at medium reasoning effort I still barely made a dent in my 5x Max plan.
I launched additional instances - a few QOL mods. Another moderate-sized expansion of Empire content, this time focusing on adding new unit types, making the Imperial court feel more alive, additional political and economic systems.
The biggest headache was that if I had the OLED TV I use as my monitor set to sleep, the pc would eventually stop rendering things. Screenshots would come out black and even worse, the game wouldn't launch. An agent found a way around the latter, though there was no way for them to change actual sleep or power settings. I didn't press that too hard, there's probably something, and I don't want to risk screen burn in.
This added a little extra time as I couldn't check some things while I had downtime at work.
I then had an agent make a LLM-powered storyteller, and a better one that any out there. I even had it figure out how to use CC instead of my API credits for the calls.
I also discovered the agents could schedule cron jobs, and later, found out that they could message each other for coordination purposes. Made my life even easier.
I set one of the spare agents to make programming art for assets (Opus is quite good at that) to save me the minimal hassle of using a dedicated image model. It also made decorations and UI elements for the other models, and seemed to be enjoying itself. In the Singularity, the most important thing is to have fun and be yourself.
I also had them make a multi-purpose bridge with hooks into the game state to ease their testing, which works just fine. Computer Use would be better, but would mean I'd have to leave the screen on and surrender control of input when I'm using the pc myself.
Later, I had one of them run performance profiling to find out what was making my rather bloated mod list slow down. It was able to figure out the culprits, including breaking things down by constant TPS tax and 1% lows or stutters.
I then had it simply fix a few mods. Easy as that. A few code optimizations, fixing bad patches.
An agent with downtime got tasked with code review. It found a few bugs, and then worked with the agents with ownership to fix it all.
Still wasn't happy with the performance, but it quickly became apparent that the only solution would be jumping to the 1.7 beta where they had swapped out the game engine from Unity to Waylith (maintaining near perfect backwards compat for mods and saves!). I'd eat my hat if a project of that scope hadn't benefited from the recent leaps in AI coding.
This went surprisingly painlessly. Most of my mods survived, and even when I had to remove several performance mods that were no longer compatible, I still got nearly double the TPS and like a fifth the load time on boot.
The agents got their mods updated without major hitches, the only one that didn't survive the transition depended on a different mod that hadn't been updated. I can probably solve that too if I try.
I then managed to get a solid game in, and the majority of bugs I found were probably to do with the original mods and not mine.
Now that is
pod-racingvibecoding. I feel like a slightly clueless but well-meaning PM with a full staff of Senior Engineers on call. I say something, it gets done, and when it doesn't, they usually find a workaround or alternative. There's no observable degradation in performance before or after compaction. I remember when we had to be meticulous in prompting, and and precisely define tasks and goals. Now, I often just half-arse things by telling Opus to use its own discretion or stick to stopping by to give feedback. My primary contribution is enough game knowledge and wider awareness to come up with ideas in the first place.I'm currently considering a mod that simply makes or sets up bases in Rimworld for me. I restart playthroughs quite regularly (mods breaking, TPS death, boredom from lack of challenge) and I find making a brand new base a chore at times, even with God Mode. Ideally, I'd like Opus to be able to just make something for me on the fly, and make the colony feel like something that's growing of its own volition, rather than helpless without my constant intervention.
I think this would work, though making it as quick and responsive as I want might be a challenge. Expanding my existing feature-complete mods wouldn't break a sweat. If Claude had native image gen capabilities, that would have been even nicer. Creating art with code works fine in some places, but it's not a general purpose solution. I can probably work up a way to get it to talk to Codex.
I'd worked with Sol 5.6 and 6.0 for my last mod, and a bit with Astra, but they were too token hungry for my taste. Opus 5.5 is simply much more pleasant to work with, and takes an appropriate but not excessive amount of initiative, while demonstrating good taste. Even Opus 4.7 and 4.8 had guzzled usage on a 20x plan, let alone a 5x one. It's a slow day at work, and I really wish I'd booted my pc up with CC running and prepped things for later.
And no, no real disasters from giving the agents unrestricted access to my C drive. Worst that's happened is one of them overwriting a mod list, or sometimes getting in each other's way. I have nothing on it that I can't afford to lose, really.
I could live like this. The bottleneck is simply my imagination, and I have plenty of imagination.
Had a very similar experience at a smaller scale over the past couple of weeks. I was tasked with producing an exercise for an enterprise IT software team that utilized some lite RPG and board game elements along with the reality of what they do.
The premise is "The Iron Triangle" of time/money/quality and how that influences up across the 4 core factions in every organization. They each have fundamentally different mandates but and ways to get there but a broadly similar goal.
I was able to produce a board game with monolithic rendering engines for web and print, rules, a pipeline to pump it on version increments to our file share, and the flexibility to add scenarios to the base game or rebrand it with the client's colors. There's a sim and strategy engine that updates balance figures.
Human playtesting was utterly necessary and seamless to integrate. I'd literally point claude code at the transcript file from our playtest and we'd run through another version iteration iteratively over 20 minutes and a couple thousand tokens. After 3 cycles it's at the level of quality I find acceptable, and the half day session with 8 participants was worth $13,000 to someone!
Beyond that it's just... fun? With the right facilitator the dry flavortext about the shit we have to deal with every day is actually kind of funny. The scenarios bring in real tension, arguments about what to do next, and a cool combination of teamwork and competition among the factions.
I cribbed a whole lot from Dune Imperium, Sycthe, Magic, and go knows what else but even as a pale imitation of what a real artist would make it was the most fun I've had at work in a long time.
13k? At API prices? Jesus. I suppose that's cheaper than testing on prod, though I would dispute that's "smaller scale".
I'm hoping you're just European and mean $13.
I've only just been surprised by the fact that Things Just Work. I make a point of having LLMs try tasks that I'm not entirely sure they're capable of, because I end up positively surprised often enough to be worth it. I was trying AI for Rimworld mods since middle 2025, and it was a slog even for a dead-simple project. It was a nuisance 6 months back. Now, the biggest concern is endless scope creep, and I can't even say that the mods breakdown when you shoot for the moon.
I would never have made these mods with AI-assistance. I know people are getting half-decent 3D models out of them, and at some point I might get even more ambitious. Depends on how the upcoming 6.2 Astra and Fable do I guess.
To be clear that's what I charged for the session, and I am using the $20 tier of Anthropic subscription. Printing the game board and collateral cost more than the AI.
Like you I've been humbled. I still see areas in which my taste and strategy are required, but the water has risen from my knees to the chest.
I could stress about it but building the things I want without endless conversations with humans has felt amazing.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link