Pietro Schirano, known for building Claude Engineer, is applying Claude Fable 5 in a targeted way that leans on its strengths in long-horizon reasoning while routing actual code generation to other models that can handle detailed specs reliably.
You should basically never use Fable for coding, but instead use it as a planner/orchestrator.
Most of today's advanced models can implement a spec perfectly, and once done you can send the work to Fable to review.
This has been my most powerful flow so far.
Early reports show developers getting stronger results when Fable handles decomposition, delegation, and final review, leaving implementation to cheaper advanced models.
At double the rate of prior flagship models, Fable's cost encourages selective use focused on orchestration rather than every coding step.
Many users endorse Fable AI as a planner and orchestrator for coding workflows rather than a direct coder because its managerial skills and systems thinking improve planning, reviewing, and orchestration efficiency.
Basically this is my flow now:
- Fable writes an in-depth plan as an md file - Send that file path to Codex with /goal
You should basically never use Fable for coding, but instead use it as a planner/orchestrator.
Most of today's advanced models can implement a spec perfectly, and once done you can send the work to Fable to review.
This has been my most powerful flow so far.
You should basically never use Fable for coding, but instead use it as a planner/orchestrator.
Most of today's advanced models can implement a spec perfectly, and once done you can send the work to Fable to review.
This has been my most powerful flow so far.
@skirano Did you get a moment where Fable reviews and is like, “you know what, I’ll do this one”?
You should basically never use Fable for coding, but instead use it as a planner/orchestrator.
Most of today's advanced models can implement a spec perfectly, and once done you can send the work to Fable to review.
This has been my most powerful flow so far.
@skirano I’m so scared to do this lol.
Part of me wants to believe in big model + composer.
But part of me wants big compute everywhere
You should basically never use Fable for coding, but instead use it as a planner/orchestrator.
Most of today's advanced models can implement a spec perfectly, and once done you can send the work to Fable to review.
This has been my most powerful flow so far.
yeah i would recommend pietro's advice lol
You should basically never use Fable for coding, but instead use it as a planner/orchestrator.
Most of today's advanced models can implement a spec perfectly, and once done you can send the work to Fable to review.
This has been my most powerful flow so far.
@skirano Don't do orchestration with it, Opus is more than capable. Depending on orchestration flow Sonnet is sometimes viable option.
For planning - 1000000% correct.
Tip of the day, let Opus also explore with N subagents + summary then ask Fable to validate findings and plan.
@Gerry It really depends on how long your work sessions are. I run loops on Codex of an average of 5/10 hours.
That would be unbelievably expensive with Fable.
@skirano Is it less work for the Principal llm (Fable) to: - investigate the issue - come up with the plan - explain in in a way that there won't be misunderstanding - then run a review prompt loop till it is correct
Hunch: It will be faster and more accurate for Fable to just do it
Even better (for bigger/complex projects) Fable writes in-depth plan Hand it to Codex for review Iterate Send to Droid with Fable In spec mode, GPT as main / worker agents Give the plan back to Claude (and optionally Codex) Give feedback back to Droid
Overkill for most things - I only do this for bigger, more complex technical work, not tasks that are smaller, more iterative, or more straightforward.
@skirano me to Sonnet 4.6: we're going to create a new /run-skill-generator for designing new software features: You should basically never use Fable for coding, but instead use it as a planner/orchestrator.
@skirano usage and goals for the week) OR if we're completely out of tokens/approaching 100% of token usage for the week divert to: Send that file path to Codex with /goal
@skirano Most of today's advanced models can implement a spec perfectly, and once done you can send the work to Fable to review. Make this the flow: - Fable writes an in-depth plan as an md file Send that file path to a new Session of Claude Opus 4.8 or Sonnet 4.6 (depending on token
@skirano I used it to build a farm. Seemed to do okay
@skirano i have been very obsessed with letting fable orchestrate codex directly.
@skirano 1/
DeepSeep V4 Pro vs Fable is $1.20 vs $60 Swarms of asynchronous teams with dedicated LLMS/skills/extensions and objectives, intercom for inter-agent comms & hierarchical supervisors, Dockerized Headroom (NixOS issue), Graphify, Understand Anything.
For me, the ultimate intelligence benchmark is still text writing. Which is kinda ironic, because models that were initially created to generate text are still often that bad at it.
If you write professionally, you quickly notice that it’s usually easier to write the text yourself than to edit whatever the model generated. Today, only the largest models can be genuinely helpful with that. That’s GPT-5.5 Pro — which, judging by its behavior, feels to me like a successor to the sunsetted 4.1 large model — and, as of today, Fable too.
My benchmark is pretty simple: I take large portions of fiction books by no-name authors, including me 😅, and ask the model to continue them. Only the larger models are really able to capture the voice, the barely noticeable nuances of vocabulary, and the author’s biases. Smaller models quickly fall back into default, flavorless narration, with a few bits of pretentiousness sprinkled in.
@skirano Agreed. Fable's managerial skills are very strong, and this is the right way to use them. Wrote an article on that specific concept and its implications on tokenomics you might enjoy:
@skirano This is golden actually
@skirano Also for recursively improving/dogfooding skills
@skirano I would sharpen this. Use Fable to write a spec doc that is actually a code implementation doc, wherein the coding agents are not forced to make decisions. A plan can be too vague.
