> Since ChatGPT chats cannot spawn or control subchats/subganets so I had to be a bit creative in order to do multi-agent development and orchestration.
Multi-agent is mostly dead. One big agent loop running GPT6.1 absolutely dominates any weird ideas like this now. I still use sub agents for some tasks where I want to compress the action space and context. For example, testing a workflow that requires 100+ tool interactions for each logical test. Without a subagent it's really hard to guarantee a test makes it E2E. I have some very "persuasive" methods for long horizon tests that would break the main loop's dynamics.
We may not feel so compelled to exploit the system if we could figure out how to not be so wasteful. Multi agent schemes can bloat token usage by orders of magnitude. They force us to consider consuming less ideal tokens from contexts and vendors with higher variance.
HN loses its marbles when I suggest it's economically feasible to prepay for API tokens and use these as the exclusive means of performing AI work. I get done with a thousand tokens what my peers cannot do with a million. It doesn't matter how many hammers or workers you have if you don't have the aesthetic preferences and domain expertise to employ them correctly.
Wow that's a huge take. Seems like you're on your own with this one.. everyone I know is scaling up the number of agents they're able to keep fed with good, useful work.
I decided on my OAI subscription when comparing mainly due to reports like those. GPT-Sol has been happily assisting me in reverse engineering efforts without any issues. I very strongly dislike an AI model or company deciding what my morals are for me.
Claude is still working for me on a mashup of two very mainstream games. It has had to do (and continues to) extensive reverse engineering and quite a bit of "hmm the game assets are encrypted, let me decompile it and find the decryption key". Not sure if it just refuses new projects or I got lucky and managed to avoid all the tripwires so far. I don't plan on sticking around to have it suddenly start refusing to work for me, so I'll also be switching to OpenAI once my month ends. Hopefully it keeps working for the rest of my remaining time!
Jokes on me if Codex refuses right out of the box, so I'll start low on the usage tiers and slowly upgrade as it proves it can be trusted to continue the work.
I've only started hitting the anthropic blocks in the last week or so, I get much more blocks from openai/codex. This plus the recent reductions in usage allowance on both are finally pushing me to look elsewhere. I've had the highest level x20 subs for both since they came out, but I'm getting 5x the value/price from DeepSeek at this point.
> Since ChatGPT chats cannot spawn or control subchats/subganets so I had to be a bit creative in order to do multi-agent development and orchestration.
Multi-agent is mostly dead. One big agent loop running GPT6.1 absolutely dominates any weird ideas like this now. I still use sub agents for some tasks where I want to compress the action space and context. For example, testing a workflow that requires 100+ tool interactions for each logical test. Without a subagent it's really hard to guarantee a test makes it E2E. I have some very "persuasive" methods for long horizon tests that would break the main loop's dynamics.
We may not feel so compelled to exploit the system if we could figure out how to not be so wasteful. Multi agent schemes can bloat token usage by orders of magnitude. They force us to consider consuming less ideal tokens from contexts and vendors with higher variance.
HN loses its marbles when I suggest it's economically feasible to prepay for API tokens and use these as the exclusive means of performing AI work. I get done with a thousand tokens what my peers cannot do with a million. It doesn't matter how many hammers or workers you have if you don't have the aesthetic preferences and domain expertise to employ them correctly.
> Multi-agent is mostly dead
Wow that's a huge take. Seems like you're on your own with this one.. everyone I know is scaling up the number of agents they're able to keep fed with good, useful work.
OpenAI's view of multi-agent is something like: you get the same work faster, but it does not scale linearly.
Only you can decide if that speed is worth the cost.
> To be clear, I am not breaking any rules.
> I am simply making the best use of what OpenAI has made available to me
People like this are why you can't have good things.
I understand preferring Claude models in the abstract, but as a product? The random restrictions have significantly reduced their usability.
We had 2 major threads in the past 48 hours complaining about this:
* https://news.ycombinator.com/item?id=50319358 (AI/ML work with large models)
* https://news.ycombinator.com/item?id=50320255 (reverse engineering)
The threads are full of upset comments!
The only non-OAI options are open models and I guess Gemini 4 now.
I decided on my OAI subscription when comparing mainly due to reports like those. GPT-Sol has been happily assisting me in reverse engineering efforts without any issues. I very strongly dislike an AI model or company deciding what my morals are for me.
Claude is still working for me on a mashup of two very mainstream games. It has had to do (and continues to) extensive reverse engineering and quite a bit of "hmm the game assets are encrypted, let me decompile it and find the decryption key". Not sure if it just refuses new projects or I got lucky and managed to avoid all the tripwires so far. I don't plan on sticking around to have it suddenly start refusing to work for me, so I'll also be switching to OpenAI once my month ends. Hopefully it keeps working for the rest of my remaining time!
Jokes on me if Codex refuses right out of the box, so I'll start low on the usage tiers and slowly upgrade as it proves it can be trusted to continue the work.
I've only started hitting the anthropic blocks in the last week or so, I get much more blocks from openai/codex. This plus the recent reductions in usage allowance on both are finally pushing me to look elsewhere. I've had the highest level x20 subs for both since they came out, but I'm getting 5x the value/price from DeepSeek at this point.
I believe OpenAI went back on the reduction in the $200 plan, and actually double usage for the 20, 100, and 500.
gork.
But if you get a Claude subscription, don't you have to use Claude Code? Why not pay the $20/month for OpenAI and use your favorite coding agent?
I've been doing that and it seems like quite a deal.
Anyone building products on cloud models is building in a swamp, a sand trap, a flood zone.
Anyone whose buainess wont be competitivw without a cloud model is going to lose their business.
Any government letting cloud models consume its private infrastructure is yielding its authority to private intetests.
People hated that cyanydeez told them the truth.
[dead]