Agents at Work: Anthropic Closes the Free Lunch, EY Bets Big on Audit Agents, and the NIST Standards Clock Is Ticking

News

EY Deploys Multi-Agent AI Across 130,000 Auditors in Its Biggest Bet on Agentic Infrastructure

EY embedded a multi-agent framework built on Microsoft Azure, Foundry, and Fabric directly into EY Canvas, its global audit platform processing "more than 1.4 trillion lines of journal entry data per year." This is one of the largest real-world agent deployments on record, targeting 100% agent-supported audit activities by 2028, and it sets a new benchmark for what enterprise-scale agentic AI looks like outside of tech.

DigitalOcean Acquires Katanemo Labs and Its Open-Source Plano Agent Data Plane

DigitalOcean bought Katanemo Labs to add its open-source Plano project, a framework-agnostic data plane that handles orchestration, observability, and safety for agentic systems, to its inference cloud. The play is clear: GPU capacity alone is not enough to win the agentic infra race, and this acquisition gives DigitalOcean a full-stack story from compute to agent lifecycle management.

Tutorial

How to Build a Self-Designing Meta-Agent That Constructs and Refines Task-Specific AI Agents

This hands-on tutorial walks through building a meta-agent that analyzes a task description, selects tools, picks a memory architecture, and spins up a fully working agent runtime without manual configuration. The self-evaluation and iterative refinement loop is what makes this worth studying: it is a blueprint for moving from hand-crafted agents to agent-generating systems.

Tools

Codenotary Launches AgentMon: Enterprise-Grade Monitoring Built for Agentic Networks

AgentMon tracks token usage, model selection, file access, secrets handling, and communication paths between agents in real time, giving security and ops teams a unified view of what their agents are actually doing. As more enterprises push agents into production, the gap between "deployed" and "observable" is where most governance failures start, and tools like this are filling that gap.

Analysis

AI Agent Governance: Runtime Enforcement for Security, Cost, and Compliance

This is a sharp breakdown of why agent governance is fundamentally different from observability: one is a camera, the other is a lock. The post covers real incidents from early 2026 and lays out the three questions every agent action should answer before it executes: Is this authorized? Is there budget? Will it be auditable?

Policy

NIST AI Agent Standards Initiative Enters Active Listening Phase in April 2026

NIST's Center for AI Standards and Innovation is now running sector-specific listening sessions in healthcare, finance, and education this month to shape standards for agent identity, authorization, and interoperability. Most enterprises will deploy agents before any of these standards finalize, which means teams building now should be mapping their controls to the emerging NIST framework or face costly retrofits later.