The Independent
Voice on AI
Control layer
Featured

AI agents can be unpredictable. Your control layer shouldn't be.

In multi-agent systems, agents can be individually correct and still produce a wrong result when they work together. One agent’s output becomes another agent’s context. Information can be lost or misinterpreted during a handoff, shared state can drift, and agents can get stuck in loops or deadlocks. When several agents rely on the same underlying model, they may even reinforce the same mistakes rather than catch them.

Subscribe to get latest news!

Deep insights for enterprise AI, data, and security leaders

By submitting your email, you agree to our Terms and Privacy Notice.

Orchestration

View all

How enterprises coordinate and scale AI in production.

image of ai harnes image

Cohere's North 2 puts AI agents on a budget and gives them a memory

Cohere is going after two common enterprise agent problems with North 2, the new version of its enterprise agent platform. It adds user quotas, rate limits, organization-wide caps and a redesigned agent harness. According to Cohere, North 2 uses memory to keep an agent's context across sessions and gives teams a shared library of knowledge and assets to draw on. It also runs in the cloud, on premises or fully air-gapped.

Infrastructure

View all

The hardware and platforms underneath enterprise AI.

Events

View all

Conferences and summits where enterprise AI strategy gets made.

Data

View all

Pipelines, architecture, and governance for AI-ready systems.

Security

View all

Protecting AI systems and the data behind them.

Technology

View all

Enterprise tech coverage through the lens of AI adoption.

Newsroom

View all

Video

View all
VB Transform 2026

Zendesk's 20 Billion Conversation Problem

Agentic infrastructure: How legacy stacks create the bottleneck to AI autonomy Multi-agent systems fail in production if the infrastructure underneath them isn't designed for autonomous, long-running loops. Token costs compound across chained tasks, latency bottlenecks surface inside orchestration layers, and agents drift from their original intent with no mechanism to catch this drift before downstream damage is done. This panel brings together agentic AI leaders who have moved past first-generation agentic deployments to deconstruct four key drivers of success: Inference Economics: How teams are managing cost-per-task in long-running agentic loops, including where speculative decoding and purpose-built small language models actually reduce spend versus where they add complexity. Physical Infrastructure decisions: Compute density, storage, memory, GPUs, inference processing units Orchestration Design: When a central master agent creates a single point of failure versus when a decentralized mesh or hybrid approach introduces coordination overhead and how to make that call based on your workload. Governance at the Infrastructure Layer: How teams are embedding rate limits, kill switches, and audit trails directly into the infrastructure stack to contain runaway API calls and prevent cascading failures before they reach production users.

VB Transform 2026

220 Million Miles. Zero Driver.

VB Transform 2026 Highlights Intelligence at Scale: How Waymo Builds Safe, Efficient AI for the Physical World The Waymo Driver represents the most mature application of AI operating in the physical world. With over 200 million fully autonomous miles driven and an expansion plan for 20+ cities, the company’s pioneering work serves as a roadmap for how to build AI systems that are reliable and safe, even in the most unpredictable scenarios. In this session, Waymo’s director of systems intelligence and machine learning, Manasi Joshi, reveals the key engineering approaches behind their operation and how data systems are improving Waymo's already exemplary safety record. This session will provide lessons for technology leaders in all industry verticals who are scaling their own agentic deployments.

VB Transform 2026

The 'Two Poisons' Problem in AI Testing

VB Transform 2026 Highlights LLM-as-a-judge vs. human-in-the-loop oversight: The great debate on model validation As agents move from prototype to production, the evaluation bottleneck has become one of the biggest hurdles. Can we truly trust an LLM to grade another LLM, or are we simply building a hall of mirrors? In this session, panelists weigh all sides of model validation, testing and putting trusted agents into production. They’ll get into the necessity of LLM-as-a-judge for real-time, automated observability at scale vs. the real world challenges of poor agent responses and the indispensable role of golden-set validation and human-in-the-loop oversight.

More