Own the Stack π§
My multi-model dev loop:
π― Plan: Claude Fable 5 high / GPT-5.5 xhigh
π Architect: GPT-5.5 xhigh
βοΈ Hard code: GPT-5.5 med / Opus 4.8 high
π Routine: Qwen3.6-35B-A3B Q6 (homelab)
π Review: cross-model - no self-grading
β Verify: Haiku/Qwen
Implement -> Review -> Verify -> Improve -> Repeat.
Own the Stack π§
My multi-model dev loop:
π― Plan: Claude Fable 5 high / GPT-5.5 xhigh
π Architect: GPT-5.5 xhigh
βοΈ Hard code: GPT-5.5 med / Opus 4.8 high
π Routine: Qwen3.6-35B-A3B Q6 (homelab)
π Review: cross-model - no self-grading
β Verify: Haiku/Qwen
Implement -> Review -> Verify -> Improve -> Repeat.
Works best with Herdr and / or tmux installed.
Gets even more comfortable with PI-Agent (less overhead).
Can easily be adjusted to Cursor or Local Models.
Codex:
https://t.co/KeGwBdmy0Y
Install as Plugins.
1: Run '$implementation-skills:prepare-work <Change / Feature>'.
2: Run '$implementation-skills:plan-slices <Instructions>'.
3: Run '$implementation-skills:implement-issues watch every 10 minutes'.
Enjoy.
Starting point for Claude:
https://t.co/lpECo36xBS
Install them as Plugins.
1: Run '/prepare-work <What you want to achieve>'.
2: Run '/plan-slices <Instructions for Implementational Slices>'.
3: Run '/loop 5m /implement-issues <Optional Instructions>'.
Enjoy.
@lanamolx@MatthewBerman@theo When running with Codex / Claude: Basically nothing additional. You would need decent hardware for running your Workloads / Applications.
But since the whole generation is handled on the cloud, a 16 GB Machine should be plenty, Processor would not matter, and storage neither.
My Fable 5 Experience so far:
- Very good planning
- Eats tokens like hell when implementing
- Utterly good problem solving
- Way better than Opus in Orchestrating (even at low)
Impressed. And I want to keep it.
Ouch.
Thank God I'm on two Max plans.
And please include Fable in the Max plans after July 7th. It would be a no-go, if I would have to pay 300-400 bucks each day for it.
Ouch.
Thank God I'm on two Max plans.
And please include Fable in the Max plans after July 7th. It would be a no-go, if I would have to pay 300-400 bucks each day for it.
The only screen in the official Claude app I have open the whole time.
Actually the only reason I have the app installed, TBH.
And thank god, I have another 2x Premium Teams subs available, and could revitalize another Max sub.
Weekend is going to be full of work.
@theo@trq212 might be a nice guy, but last time I checked, he is not the official spokesperson of Anthropic. Until I see it in my sub (or get offered to sign up for $500-$1k sub), I simply don't believe anything I am told.
Sorry, I don't believe that.
Why would you guys communicate the fact of "some tasks like coding and debugging" being routed to Opus, if you would not have implemented guardrails making the thing basically unusable?
And BTW: No word on that via the official accounts.
No offense.