AI Briefing
KOSign in

How Balyasny Asset Management Evaluates and Governs Claude Fable 5 for Agentic Workflows

·2026.09.17 09:00

Key point

Balyasny Asset Management uses Claude Fable 5 for agentic workflows, achieving an 89.4% success rate on financial tasks and reducing merger-arbitrage analysis from days to under a day.

Details

Balyasny Asset Management (BAM), a global investment firm managing roughly $38 billion in assets, has integrated Claude Fable 5 into its core workflows to enable autonomous, multi-step financial analysis. Charlie Flanagan, BAM’s Chief AI Officer, describes 2026 as the year AI shifted from search to work, driven by harnesses like Claude Code that allow agents to sustain long-running tasks until completion.

Evaluation and Performance

BAM evaluates new models on thousands of real-world financial tasks with verifiable outcomes across equities, macro, and commodities. On this relevant subset, Claude Fable 5 achieved an 89.4% success rate, compared to 86.1% for the prior production model. The model notably succeeded on a set of economics problems that no previous model had completed, a result BAM verified through independent checks and review with Anthropic.

Operational Impact

The deployment has significantly reduced time-to-insight for complex workflows:

  • Merger-Arbitrage Analysis: Tasks that previously took three to five days now take less than one day, with agents running for approximately 30 minutes to build deal-analysis packages.
  • Tax-Loss Harvesting: An agent explored 90,000 database tables to build a weighting system, producing results more comprehensive than traditional approaches.
  • Central-Bank Analysis: A recurring analysis for the Chief Economist was reduced from roughly two days to approximately 30 minutes.

Governance and Infrastructure

BAM built BAMAgent, an internal platform supporting thousands of autonomous agents working 24/7, to securely deploy these capabilities. The firm treats safety as an operating-model question, enforcing strict controls including approved data boundaries, least-privilege access, and human review for material outputs. While Claude Fable 5 handles planning and reasoning, BAM’s in-house harness manages tool access, permissions, and monitoring, ensuring that investment judgment and accountability remain with human professionals.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.