It doesn't work. It's far too expensive. The orchestrator is where the token cost is on every turn. So if you're just getting Fable to plan and then task off other agents that are various models to go and implement, you're still having to read every token at every turn at a Fable 5 rate. You need to be able to route the orchestrator based on the task so Fable 5 is an epistodic planner and then Opus is a steady-state orchestrator.
2026-07-04 12:24:00
0
abe22222334 :
The explore agents are gonna leave out good stuff they condense and forget too much context the loss is to great and since you don’t have the best and all information it’s gonna suck and forget stuff and mess something up fable and all of Claude doesn’t know anything that happened after Jan 1 2026 unless you tell him or make him the agent research said thing so even if you have him make something and check it your still not even up to date and people will be ahead of you and you building something already built or in the past even though you think your not bad video
2026-07-03 13:25:42
2
Maximus :
rofl opus 4.8 for review 😂
2026-07-02 22:34:32
0
Leibo :
You never disappoint, I incorporate multi-runtime evals for reviews but this doesn’t work inside of dynamic workflow loops.
2026-07-03 00:58:58
1
Nikhil Tulshigar :
use tokenlayer.net to use fable for free
2026-07-11 09:53:19
0
JJ | AI in a Bottle :
interesting! Thanks for sharing
2026-07-03 19:56:13
0
Rex Allen :
cc keeps telling me graphify is not beneficial to my projects
2026-07-03 06:20:06
0
Nate Auchter :
Brilliant
2026-07-03 05:40:56
0
mr_bix :
Autonomy
2026-07-03 10:26:32
0
Galactic Star Man :
only can do 1 run even on max plan so it is not usable
2026-07-02 18:20:53
0
abe22222334 :
Bad
2026-07-03 13:22:52
0
p :
Bro you making a rocket?
2026-07-02 23:04:23
5
A05_BCA :
Well give her a shot I’ll run it tonight
2026-07-02 17:23:54
0
Bill E. Goat🐐 :
Instructions unclear .. spawned 1000 agents after planning to execute, zapped tokens.
2026-07-02 21:03:50
0
To see more videos from user @agentic.james, please go to the Tikwm
homepage.