I ran an experiment to see how far it could get with a top-down 2d game, like a more challenging version of "draw a pelican." I'm waiting on Fable to rewrite the whole thing now, but I was impressed by how far Opus 4.8 got with it: https://github.com/jmtame/scrapland
Started out as a one-shot attempt, but ~200 prompts later it's at a place where it's at least fun to watch the AI teams destroy each other.
Yep that’s right. First sentence was poorly worded, but I was trying to figure out why moving to Austin made me feel nostalgia, turns out it was the Cumulus clouds. Both Austin and Illinois have them, but the Bay does not.
When I moved from the Bay Area to Austin, the first thing I realized: I missed seeing Cumulus clouds, which I saw a lot growing up in the midwest. Bay Area either has blue skies or Cirrus clouds, but never did see Cumulus clouds there.
Thanks for sharing this. Going to try it out on a game inspired by Rust. It's helpful re: the point on rodney - I've had a hard time getting the testing to work well in the browser.
This comment should be higher given how much FUD is in these comments. Hacker News having a meltdown about a bunch of backoffice folks getting laid off while they’re still aggressively hiring SWEs is… something
I remember my first few weeks of Claude Code. The high will wear off as you bump into the limitations, and then it starts to feel like you're more of a "manager of a junior-ish dev." The work shifts to clarity of intent and capturing edge cases, rather than purely coding. It's a fun time when you first jump in, but don't be surprised when your excitement reverts back to baseline.
Fair enough, I tend to avoid overly negative people. Criticism can come from a good place to suggest improvements (radical candor), but agreed that some of the comments are just personal attacks. I think we're both in agreement that those aren't people we'd want to work with.
Nice, nothing like a little personal bias to inject into an interview process. If you can't handle criticism and you're just looking for sycophants, you're probably not the type of employer or hiring manager most people want to work for anyway.
The origin story of Rust is classic: they got tired of Day Z and wanted to make it better, so they hired some random contractor to copy Day Z with elements of Fortnite and Minecraft, the developer complains that he's entitled to more money from the success of the game, lawsuit follows, and then Facepunch supposedly claims the original version was so buggy they have to rewrite the whole thing from scratch. Unclear if they were just trying to start fresh so the lawsuit wouldn't follow them forever, but it started out as just a clone of another game and they turned it into a hugely successful business (and an incredible game). Most games have a short shelf life, but I've watched at least over the past 4-ish years, and the rate that they continue to push changes is impressive.
Rust has so many compelling features as a game. It was the first game where I felt like I thought about it while I wasn't playing it, because your character remains "in the game" and your base can be raided even if you log off. I don't play a ton of online games, but that was a very new and different concept when I first discovered Rust.
The game reminds me of sitting down at a poker table in a casino. It's very unforgiving - you grind, invest a lot of time, and make calculated bets as to whether you can win or lose a raid, but you can instantly lose everything in a failed raid.
I wish someone would make a browser-based version that was fun to play, and I've thought about it for some time, but the struggle is scoping an MVP that is as compelling given the constraints (eg a 2d or top-down version makes it harder to do things like build multi-story buildings and raid them).
Not to mention how much worse it is on mobile. Every web site asks me to accept their cookies, close layers of ads with tiny buttons, and loads slowly with ads spread throughout the content. And that’s just to figure out if I’m even on the right page.
You’re referencing an old and outdated technology that has no capability to handle things like revenue and attribution. New protocols will need to evolve to the current use. Owners want money, so make the protocol focused on that use case.
I’m not here to propose a solution. I’m here as an end-user saying I won’t go back to the old experience which is outdated and broken.
Cloudflare is a gatekeeper because they’re trying to insert themselves between the owner and the end-user. Despite all the altruistic signaling, they really just want to capitalize on AI. And they’re happy to do that even if it results in a subpar experience for the end-user. They started this with a focus on news organizations, so I’m not particularly excited about trying to block AI access and lock down the web through one private company just so we can preserve 90s era clickbait businesses.
I pretty much use Perplexity exclusively at this point, instead of Google. I'd rather just get my questions answered than navigate all of the ads and slowness that Google provides. I'm fine with paying a small monthly fee, but I don't want Cloudflare being the gatekeeper.
Perhaps a way to serve ads through the agents would be good enough. I'd prefer that to be some open protocol than controlled by a company.
I didn’t realize screenshots worked until a few days in, that was a great discovery. And recently learned you can directly paste in copied screenshots using ctrl+v (instead of cmd+v on a Mac).
I've thought about this lately. In order to do that, you need to know where people typically stumble, and then create a rubric around that. Here are some things I'd look for:
- Ability to clearly define requirements up front (the equivalent mistake in coding interviews is to start by coding, rather than asking questions and understanding the problem + solution 100% before writing a single line of code). This might be the majority of the interview.
- Ability to anticipate where the LLM will make mistakes. See if they use perplexity/context7 for example. Relying solely on the LLM's training data is a mistake.
- A familiarity with how to parallelize work and when that's useful vs not. Do they understand how to use something like worktrees, multiple repos, or docker to split up the work?
- Uses tests (including end-to-end and visual testing)
- Can they actually deliver a working feature/product within a reasonable amount of time?
- Is the final result looking like AI slop, or is it actually performant, maintainable (by both humans and new context windows), well-designed, and follows best practices?
- Are they able to work effectively within a large codebase? (this depends on what stage you're in; if you're a larger company, this is important, but if you're a startup, you probably want the 0->1 type of interview)
- What sort of tools are they using? I'd give more weight if someone was using Claude Code, because that's just the best tool for the job. And if they're just doing the trendy thing like using Claude Agents, I'd subtract points.
- How efficient did they use the AI? Did they just churn through tokens? Did they use the right model given the task complexity?
Started out as a one-shot attempt, but ~200 prompts later it's at a place where it's at least fun to watch the AI teams destroy each other.