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.
Users praise Fable AI for excelling as a planner and orchestrator rather than a direct coder because it leverages its strengths in systems thinking, review, and workflow coordination to produce smarter results.
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.
@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
@skirano i have been very obsessed with letting fable orchestrate codex directly.
@skirano This is golden actually
@skirano 100%. fable for review, not write. game-changer.
@skirano This is such a smart workflow. Using Fable as the high-level strategist/planner plus reviewer instead of the coder makes total sense it plays to its strengths without burning it on implementation details. Have you noticed a big jump in output quality or fewer revisions ?
@skirano That is a fair point actually. My longest goal was 12 hours, but that is not typical, mostly because I don't want to verify 12 hours of work... but i guess if Fable can do it... and I can trust it.... 🤔
Good point
Yep. Have found it to be great at deep systems thinking, understanding the full picture and correcting and writing specs. It actually raises all other models as well, because as you mentioned they follow a spec flawlessly, and all decision traces that may come up have been handled.
@skirano Great tip!
@skirano Exactly right. Fable has this biggest-model smell and god's-eye view... but it's not infallible. I needed 4 round -- four! -- to nail down the plan. After two rounds, Gemini-3-1-Pro thought it was all good, but Codex 5.5 went another 2 rounds with Fable until it was really done.
@skirano I wonder how many tokens you save by having Fable write a detailed spec instead of just letting it code it
@skirano I have seen when I used Fable for a goal, it mostly used Sonnet 4.6 and sometimes even Haiku for some sub-agents.
@skirano This is what i do ;)
Using ultracode to plan and then branch, worktree and codex it out.
@skirano I’ve been having it act as a manager agent and deploy sonnet sub agents to do the work but it still cranks insane tokens. I think I literally just need to use Fable to write the spec then use sonnet to execute and not use fable as a manager
@skirano What's the typical spec size you're sending to Fable for review after implementation, and how does that impact review time?
@skirano @hq_fyi
@skirano hmmm I was thinking the opposite
@skirano great callout. gonna steal this for https://github.com/Metabuilder-Labs/tokenjam
@skirano Planner + orchestrator, not coder — this is the right mental model for agentic AI workflows. Let the implementation models do the heavy lifting, use the reasoning layer to architect and review. Separation of concerns applies to AI too. 🧠 #AI #AIAgents #Dev #BuildInPublic
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.