Monday, June 9, 2025

Start of archive·May 16

Top story

The ReadVercel Blog

Building secure AI agents

As AI agents move into production, prompt injection has become the critical security vulnerability that most teams aren't designing for. This isn't a future problem—it's a build-time decision that separates secure deployments from breached ones.

Prompt injection identified as most critical AI agent security risk

The briefs

As generative AI becomes mainstream, the industry lacks agreed-upon metrics to measure success. This definitional gap reveals deeper questions about what AI adoption actually means for businesses—and why your growth metrics might be misleading.

Core problem: AI industry lacks standardized measurement frameworks

Vercel's AI Gateway observability layer removes the black box from multi-model LLM deployments, letting builders optimize for latency, cost, and performance across ~100 models in real time. This is infrastructure-as-a-feature for the app-layer.

Vercel AI Gateway supports ~100 models

Vercel is expanding v0's model capabilities on its API layer, enabling developers to build AI-powered products with longer context windows. This matters for founders scaling AI features without managing multiple model providers.

v0-1.5-md: 128K token context now in beta