The 85% Beyond the Model
summary
Claim: the agent layer is about 15% of the work and the remaining 85% is identity, permissions, deterministic decision systems, retrieval-by-tool and auditable logs. Arta Finance runs this in production across regulated markets, enforcing suitability and accreditation checks dynamically, never letting an LLM generate its own numbers, and logging every tool call and human override. Deployment uses synthetic personas, live release testing and LLM judges grading thousands of sessions per release so failures surface early and conversations are fully replayable for compliance.
AI-written summary · about
chapters (7)
- 0:00Introduction and context
- 2:58The 85% advantage
- 5:49Architecture over prompt engineering
- 7:57Structural constraints and permissions
- 10:29Evaluation systems and testing
- 11:34The backbone of wealth management
- 13:18Summary and conclusion
related talks
25mFrom Chaos to Control: Building Enterprise-Grade LLM Agents That Actually Work
44mProduction-Grade LLMs & AI Agents
17m12-Factor Agents: Patterns of reliable LLM applications
26mAI Agents as Financial Co-Pilots | AI-Driven Strategy & Decision Making
36mLLMs and AI agents demystified
29mBuilding Trust in AI Agents
01h 04mPanel Discussion: Money Moves at Machine Speed - Trust, Risk, and Quality in Agentic Finance
13mHarmonising LLM Agents for Real-World Market Intelligence:[...]