32 Agents in parallel local! 🤯 The key, you can run multiple models once they get small enough! The crazy part is that by using an improvement loop to work out what the worst model failed on, I could increase accuracy rapidy! Many possibilities if you break the task down!

Stop worrying about tokens, and get the Claude Max for $200, and the Codex $200 plan...for $400 per month, U'll get near infinite parallel agents. Stop playing around in the sandbox with API's, cursor, OSS, Local, and get building with SOTA, FAST!

Too many think AI coding is just a crazy one shot...but it's more iteration of dozens or hundreds or thousands of times! We are getting better at doing more in one go, but there is drift from the target the longer the agent works. But AI's are good at reviewing and resolving that drift, plus refactoring if it's a code quality issue. Too many stuck in 2024 with their ego, even super smart and experienced people (especially actually). Still looking for single examples of failure....well ding ding ding, it takes perseverance to push through and find things that work. Creativity. You wouldn't believe what just a slight re-wording did to something we were working on. Stop looking for confirmation bias, and build something amazing! Multiply your experience, the time is NOW to ship what you dreamed of! Up the stakes, do something EPIC, keep it challenging.

We have been creating games with AI for 1.5 years now, and MAKING a game vs Making a FUN game, is a very different prospect. Reality is the amount of games will go up exponentially, but 99% of them will likely not be that 'fun'. But obviously it will still increase the quantity of fun games. Over time, it will get easier to make fun games, the systems, mechanics, balance, and flow will be systemised, so the bar will rise. Playing the same game with your friends has benefits, and if everyone is more split up, that could be less than ideal. There will still be a need for shared games and popular ones that people discuss. Alternatively, it could get to the point where you can just 1 shot your perfect game, and everyone gets split up...dystopian, but possible. It will 100% happen in video/audio related things for sure...games a bit harder, but possible. At least for 2026, experienced game devs who embrace AI will be the winners, and the more games you make, the more experience you get. You can speed run it, if you try. John Romero made ~30 games before Doom I think! So far I've released ~10 games in 1.5 years, some BIG complex ones, that would not be possible without AI. It's only in the last few months that I feel like I am breaking through and figuring out what a 'fun' game is though.

I think I have started to figure it out. The amount of times I got raged at or laughed/ganged up on for using AI to code, especially with games is crazy. Of course, back in 2024, there was less evidence, even though I was making my own game + engine with it. However, many have tied their identity to being 'coders'. They love complexity, they love discussing code. Just look at some random discord channels in game frameworks, u see people discuss obscure syntax and techniques...rather than actually making a game or shipping. Their github is like the desert, and they have a game in progress for 10 years. As the AI to analyse their repos, and it tears them to shreds. Their commits are 'F***' , 'working' 'actually working' 'i am a genius' 'i am actually an idiot' 'refactor entire codebase to XYZ language'. Some of these guys are very smart, and are onto some stuff that makes sense...and given their knowledge if they just embraced agentic engineering and used their own context and knowledge to boost their results, they could finally fulfil their own dreams. But reality is...they are FROZEN...by "The Resistance"...from Steven Pressfield's book "The War of Art". They are now so paralysed by shipping 'good', 'performant' 'complex' 'beautiful' code, that they now just don't ship...they just talk about code, write blogs, host podcasts....and never get the reps in by actually shipping the working code. That is not even talking about making a good product or game...which requires failure mostly...John Romero's 30th commercial game was Doom, and that doesn't even count his non commercial ones. They will mock you for having 7 levels of nested IFs (I agree not idea)...but ignore the fact that Celeste and Rouge Legacy have 9-11 nested, yet were smash hits! Their ego also comes into play, if the AI can write better than them, their whole identity is challenged, they would rather 'Ignorance Is Bliss', and make themselves believe that the AI can't write good code, let alone working code...to protect themselves. They group together, mock it, and all self delude themselves. It's a sad state, and I wish I could help them, but most are just not open to it, and I can't imagine what they will be thinking by the end of the year.

Add something like this to your top level .claude/Claude.md , never worry about git commits/PR creation again! P.S I def need to clean it up, too many tokens, but it's been working beautifully.

Pixel Art ^_^