Revenue Was Down 2%.

2026-07-12 · Source: Towards AI - Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Cybersecurity & Data Privacy · Depth: Advanced, medium

Summary

An analysis highlights that slow, manual data orchestration, not reporting, prevents businesses from quickly understanding critical metric changes like a "2% revenue drop." The problem stems from human-orchestrated data retrieval across seven systems by four people over three weeks. To address this, an agent-based architecture is proposed, designed for enterprise environments, particularly regulated ones. Key design principles include ensuring the data platform produces facts while the model interprets them, querying certified metrics for traceability, and implementing a human approval gate for any write actions. The architecture leverages Microsoft technologies like Entra for identity-based access control and Foundry IQ for unified data retrieval, emphasizing condensed specialist findings over raw data to maintain performance. The author stresses that the semantic layer is a prerequisite, and the core challenges lie in data governance, identity management, and auditability, rather than the choice of AI model.

Key takeaway

For AI Architects designing agent systems for enterprise data, prioritize robust data governance and identity management over model selection. Your architecture must enforce granular access via user tokens, not shared service accounts, and include human approval for all write actions. Focus on building a strong semantic layer and ensuring agents return condensed, traceable findings to prevent "confidently wrong" answers and ensure auditability, which are critical for security and regulatory compliance.

Key insights

The core problem is human-orchestrated data silos, solvable by an agent architecture that queries certified metrics.

Principles

• The data platform produces facts; the model interprets them.
• Agents must never compute figures, only query certified metrics.
• Identity-based access control is paramount for security.

Method

Design an agent system with a supervisor and specialists, using identity-based access, a human approval gate for writes, and a semantic layer for certified metrics.

In practice

• Implement Entra "on-behalf-of" for agent identity.
• Gate all agent write paths with human approval.
• Ensure specialists return condensed findings (1-2K tokens).

Topics

• Agent-based Systems
• Enterprise Architecture
• Data Governance
• Microsoft Fabric
• Identity Management
• Multi-Agent Orchestration

Best for: AI Architect, Director of AI/ML, MLOps Engineer

Related on AIssential

• A Strong Suite in Agentic Strategy: Data Apps — AI agent success in enterprises depends on architecting governed, queryable data applications, not just advanced models or semantic layers.
• Every Agent Needs a Box — Aaron Levie, Box — Secure, sandboxed environments are crucial for AI agents to operate effectively and safely within enterprise data systems.
• The AI agent bottleneck isn't model performance — it's permissions — The core bottleneck for enterprise AI agents is permissioning and governance, not model performance.
• AI agents are running hospital records and factory inspections. Enterprise IAM was never built for them. — Identity governance, not model capability, is the primary barrier preventing agentic AI from moving from pilot to production.
• The Complete Architecture for Write-Capable AI Agents: Data and Security — Write-capable AI agents demand robust architectural safeguards for data integrity and security, beyond just model performance.
• Many Companies Use AI. Few Know How to Build an AI-Native Enterprise Data Platform. — AI agents transform enterprise data ecosystems by autonomously executing multi-step tasks, moving beyond simple chatbot interactions.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Towards AI - Medium.