Matthew Cosier

Matt lives in Melbourne Australia. An experienced .NET developer and serial entrepreneur, his latest venture: Portfolio Cloud #fintech

[Matthew Cosier @cosier]

@iruletheworldmo I want a discount on my next punnet.

[Matthew Cosier @cosier]

@theo claude.md is a one liner telling it to read agents. md :) (or vise versa depending on your pain threshold)

[Matthew Cosier @cosier]

@weswinder Xhigh fast, my temporary friend…

[Matthew Cosier @cosier]

@haider1 Why does GPT pro always get left out of these discussions and boxed in with the agentic models

[Matthew Cosier @cosier]

@chaosengineerr Because they get paid $2M a month

[Matthew Cosier @cosier]

Bye bye miss Extra High, it’s the 5th of the month and 10x credits will die … thanks for the ride @sama

[Matthew Cosier @cosier]

@ClaudeDevs My token balance is looking gre …. Annnndddd it’s gone.

[Matthew Cosier @cosier]

@reach_vb @Kappaemme1926 It’s faster than it’s ever been for me right now

[Matthew Cosier @cosier]

What on earth is up with codex right now? It’s blazing fast! Wow

[Matthew Cosier @cosier]

Anyone know how you can pin codex projects to different plugin/integrations? Which mf only has ONE linear tenant? I have like 10..

[Matthew Cosier @cosier]

#olddog

[Matthew Cosier @cosier]

@thsottiaux @sama can you confirm a 5.6 party? I think we need one. And I’ll happily not go again, if you can provide me with another extension on my 10x rate limits :) 🙏

[Matthew Cosier @cosier]

@Jmoon_174 Agreed! For now our best and most optimized workflow is leaf node execution / offloading with explicit / highly focused prompts to get the most out of it. GPT lives on as the orchestrator for now. Was probably always the right set up anyway.

[Matthew Cosier @cosier]

Opus 4.8 is bending my mind. I’m convinced it’s the CC harness in complex codebases with multi agent orchestration that is fried. Single shot documentation generation type stuff is about 30-40% better than 4.7 - everything else fails miserably

[Matthew Cosier @cosier]

@argofowl Soon as a new model is incoming, they preserve compute for some reason… not sure they do it on purpose to make the new one feel more intelligent….Real reason is likely due to requiring the extra capacity for pre release testing… temporary pain for major gain!

[Matthew Cosier @cosier]

@brockpierson Yeah but I wrote my own native win32 IRC client as a bored teenager in C++ .. those were the days

[Matthew Cosier @cosier]

@AdamRackis Yeah it’s a bug unfortunately. It’s only started doing this for me since the last update - even with full permissions, on various tool calls it pauses to ask for permission - super annoying. @OpenAIDevs

[Matthew Cosier @cosier]

@mark_k 💯.. I don’t mind if it’s expensive tho so long as there is a plan that is proportionately scaled to that cost, allowing it to be consumed under realistic workloads without hitting limits. Fast & Token sufficient == good product. OpenAI gets this. Anthropic don’t seem to (yet)

[Matthew Cosier @cosier]

OpenAI is about to execute the biggest rug pull in recent time by drastically reducing our limits 😢 the only way this becomes bearable is if they give us a 50-100x plan at some significant $ bump. I’d pay it in a heartbeat

[Matthew Cosier @cosier]

What the actual fuck. "The truth is now plain and I owe you a straight account — the channel was silently
eating my writes" ..... man, Claude Opus 4.8 in CC is absolutely *fucked*