Architect unattended 8h+ agentic workflows that decompose complex software tasks without human intervention or infinite token spend.
Every lab runs completely client-side with copyable prompts and browser checkpoints.
Curriculum Progress
0 of 4 modules completed (0%)
Long-running agents run unattended for 8 to 17+ hours on ambiguous, multi-step engineering challenges. However, Ramp spending telemetry across 70,000 companies reveals that flagship models like Fable 5.1 carry only ~6% of tokens and 11% of enterprise spend—most work is efficiently handled by Opus 5.5 or Sonnet 5.5. Master the two-check framework before committing hours of unattended execution.
# Run the two-check evaluation for this workload: Task: Multi-repo dependency audit & refactor across 12 services Check 1 (Workload): Is the solution spec fully known? (No -> Candidate for Fable 5.1 / Opus 5.5) Check 2 (Price): At 15 turns and 400k cumulative tokens, what is the cost difference? - Fable 5.1: $10.00/1M input, $50.00/1M output, $0.25/1M cache - Opus 5.5: $4.00/1M input, $20.00/1M output, $0.20/1M cache - Sonnet 5.5: $2.00/1M input, $10.00/1M output, $0.10/1M cache Run /tools/agent-cost to verify cost ceilings.
According to real-world enterprise spending benchmarks (e.g. Ramp corporate data), what role does the most expensive flagship model (Fable 5.1) typically play in agentic pipelines?
Model multi-step agents with tool calls, retries, and monthly run volume.
Price an agentic coding session: turns, context, tool results, cache hit rate, and output tokens.
Simulate daily, monthly, and annual LLM bills across every model in the catalog.
Compare 100% frontier traffic against a light → general → complex model cascade.
Claude Fable 5.1 and Claude Opus 5.5 are engineered specifically for multi-hour reasoning and deep codebase orientation. Sonnet 5.5 serves as an efficient high-speed worker for bounded tasks.
Set explicit max_tokens ceilings per turn, establish coordinator-subagent bounds with isolated contexts, utilize prompt caching, and enforce spend circuit-breakers in host hooks.
Fable 5.1 ($10/$50 per 1M) is optimized for highly ambiguous multi-step tasks requiring deep adaptive thinking. Opus 5.5 ($4/$20 per 1M) offers frontier coding intelligence at less than half the cost for production execution.