← sources

Built an entire AI dev squad for my game and now im spending more time managing AIs than coding

source: reddit (reddit:ClaudeCode)published 2026-10-21 06:26ingested 2026-10-22 02:42

retained text

So im a solo dev building a tactical Cold War wargame set in West Germany 1985

Somehow ended up assembling an entire AI dev team and now idk if I’ve overcomplicated the whole damn thing

My current squad  
Me - Head developer  
Claude Fable 5.1 - Lead game designer  
Claude Opus 5.5 - UI/UX  
Kimi - Game artist  
ChatGPT Astra - Architecture oversight  
Claude Code Opus 5.5 - Head of engineering  
Claude and Codex cloud sessions - Coding workers  
DeepSeek - Coding via Hermes and Discord  
Qwen local - Another coder via Hermes  
Orinth - Currently testing and so far pretty good

Dropped Jules/Gemini because it was pretty useless for coding in my experience

DeepSeek on the other hand has been my most tireless worker. Just keeps coding away and gets shit done without costing a fortune

Both Claude and ChatGPT can access my local repo and spawn multiple cloud sessions

**Now heres the ridiculous part**  
I went from Claude Pro to Max x100 then x200 because I kept hitting usage limits. Somehow even maxed out x200 yesterday lol

So I gave ChatGPT Astra a shot at managing engineering. Initially it was alright until suddenly it couldnt get new cloud sessions to access the shell or Github repo  
Normally Claude would just go into my PC investigate and try to fix the problem

But Astra instead did some surface level checking and started advising Sol (my other ChatGPT agent managing engineering) on methods and philosophy rather than actually solving the damn issue

After 4 rounds of this nonsense it basically told me to open a support ticket with OpenAI  
Like seriously wtf

Ended up spending my remaining $5 Claude API credit to get Opus 5.5 to investigate it directly and it solved the problem in about 10 minutes

To be fair Codex itself is actually pretty decent at coding. Its the managing troubleshooting and orchestration side that really sucks from my experience

Now im using Codex on Pro while waiting for my Claude Max x200 limit to reset

The whole idea was to have multiple agents coding in parallel even overnight while im asleep or at work  
But now half the time they’re reading markdown files writing reports and managing each other instead of actually coding

**So how the hell do I run this more efficently without burning through money and usage limits**

Should I ditch the whole management hierarchy and just have one strong orchestrator with cheaper coding agents  
Anyone running a similar multi agent setup successfully especially with cheap APIs or local models

My PC has an RTX 5080 16GB and 64GB RAM so local coding is also an option

At this point idk if im developing a game or running an AI bureaucracy

document fetched 2026-10-22 00:43 · canonical https://reddit.com/r/ClaudeCode/comments/1xbfm2k/built_an_entire_ai_dev_squad_for_my_game_and_now