Opus 4.5 is not the normal AI agent experience that I have had thus far

I'll write the code, it can help me explore options, find potential problems and suggest tests, but I'll write the code.

I then proceeded to use it to hack on its own codebase, and close a bunch of issues in a repository that I maintain (https://github.com/nuprl/MultiPL-E/commits/main/).

All the LLM coded projects I've seen shared so far[1] have been tech toys though. I've watched things pop up on my twitter feed (usually games related), then quietly go off air before reaching a gold release (I manually keep up to date with what I've found, so it's not the algorithm).

I find this all very interesting: LLMs dont change the fundamental drives needed to build successful products. I feel like I'm observing the TikTokification of software development. I dont know why people aren't finishing. Maybe they stop when the "real work" kicks in. Or maybe they hit the limits of what LLMs can do (so far). Maybe they jump to the next idea to keep chasing the rush.

Acquiring context requires real work, and I dont see a way forward to automating that away. And to be clear, context is human needs; i.e. the reasons why someone will use your product. In the game development world, it's very difficult to overstate how much work needs to be done to create a smooth, enjoyable experience for the player.

While anyone may be able to create a suite of apps in a weekend, I think very few of them will have the patience and time to maintain them (just like software development before LLMs! i.e. Linux, open source software, etc.).

[1] yes, selection bias. There are A LOT of AI devs just marketing their LLMs. Also it's DEFINITELY too early to be certain. Take everything Im saying with a one pound grain of salt.

I read If Anyone Builds It Everyone Dies over the break. The basic premise was that we can't "align" AI so when we turn it loose in an agent loop what it produces isn't necessarily what we want. It may be on the surface, to appease us and pass a cursory inspection, but it could embed other stuff according to other goals.

On the whole, I found it a little silly and implausible, but I'm second guessing parts of that response now that I'm seeing more people (this post, the Gas Town thing on the front page earlier) go all-in on vibe coding. There is likely to be a large body of running software out there that will be created by agents and never inspected by humans.

I think a more plausible failure mode in the near future (next year or two) is something more like a "worm". Someone building an agent with the explicit instructions to try to replicate itself. Opus 4.5 and GPT 5.2 are good enough that in an agent loop they could pretty thoroughly investigate any system they land on, and try to use a few ways to propagate their agent wrapper.

You can see some of the context limits here:

https://models.dev/

If you want the full capability, use the API and use something like opencode. You will find that a single PR can easily rack up 3 digits of consumption costs.