Fable on Subscriptions Depends on Compute Capacity
Restoring Fable as a standard part of subscription plans is a compute capacity problem. The team is actively working on it but can't give a firm timeline.
Restoring Fable as a standard part of subscription plans is a compute capacity problem. The team is actively working on it but can't give a firm timeline.
Clarifying the exact end of Fable's availability on subscription plans: access cuts off at 11:59:59 PM PT on July 7th.
Skills are still recommended with Fable despite it being a more capable model. The main advice: most users need to make their skills shorter.
In response to a user asking for the exact cutoff time, the team confirmed subscription access to Fable ends at 11:59:59 PM PT on July 7th.
The keynote talk 'A Field Guide to Fable,' delivered at AI Engineer World Fair, is now available to watch on YouTube.
The AI Engineer World Fair keynote on Fable was based on a series of articles. The first of (hopefully) three is now live.
Responding to a thread about AI-coding burnout, the preference here is for focused single-agent sessions on hard tasks, where it's easier to catch bad assumptions and stay in the weeds. Running tens of agents in parallel is common too, but the single-agent mode has its own value for quality.
When running many parallel agents via the claude tag, asking Claude to write regular summaries across all running agents works like a chief-of-staff briefing. Iterating on the report format toward higher information density and more actionable outputs has been a useful refinement.
The dynamic workflows feature lets Claude Code spin up hundreds of subagents in parallel, each handling one unit of work (like researching a single candidate), then merges the results. The workflow is written in natural language, with a prompt specifying the role and output requirements.
Workflows for multi-agent tasks are easy enough to regenerate each time that saving them isn't worth the overhead. Generating fresh on demand is the preferred approach.