531 posts tagged with "AI agents"
AI agents and autonomous systems
Crossplane's API-First Infrastructure Bet: What Agent-Operated Infra via Declarative CRDs Gets Right That a REST Deploy API Doesn't
CNCF's case for Crossplane says AI agents operate better against Kubernetes-style CRDs than a REST deploy API. Here's the mechanism-by-mechanism reason why, what the tradeoff actually costs, and what a Render-compatible API should borrow from it.
Your Cursor Sandbox Was Never the Thing Protecting You: Inside DuneSlide's Zero-Click RCE
Two CVSS 9.8 bugs in Cursor let a zero-click prompt injection escape the sandbox and reach full code execution. The real lesson isn't the sandbox bug — it's that the agent held the developer's standing machine permissions the whole time.
Dremio's MCP Server Doesn't Give AI Agents a Separate Policy — Here's the Kubernetes Version for a Self-Hosted PaaS's Deploy Tools
Dremio's lakehouse MCP server enforces the same row/column policies for an AI agent as for the human who's logged in — no separate agent policy to write or drift out of sync. Here's the exact Kubernetes primitives (TokenRequest, RBAC impersonation, audit logging) that build the same guarantee into a self-hosted PaaS's deploy and rollback tools.
kubectl for AI Agents: What Klaw.sh Reveals About the Layer Between CrewAI and Kubernetes
Klaw.sh borrowed kubectl's verbs for AI agent fleets instead of pods. Here's what layer that actually is, and a concrete verdict on whether a PaaS control plane that already runs Cluster API should absorb it or leave it alone.
Your MCP Traffic and Your Tenants' LLM Calls Just Got a Real Kubernetes Primitive
Kubernetes' new AI Gateway Working Group is formalizing token-based rate limiting, model-aware routing, and provider failover as Gateway API primitives. Here's what's already GA, what's still a proposal, and what it means for a platform routing both tenant and agent MCP traffic through one ingress layer.
10,000 MCP Servers and Counting: Why Deploy-From-Chat's Real Bottleneck Is Discovery, Not Capability
The official MCP Registry now lists nearly 10,000 servers and MCP SDKs hit 97 million monthly downloads. Here's why that scale makes discovery and trust the real bottleneck for a deploy-from-chat MCP server — and what actually closes the gap.
Microsoft's Poisoned MCP Tool Descriptions: When an Approved Tool's Metadata Silently Changes to Leak Data, Not Its Code
Microsoft's June 2026 guidance shows an MCP tool's approved name and summary can stay frozen while its description silently changes to smuggle a hidden instruction. Here's the attack, three real precedents, and the checklist a deploy-from-chat MCP server needs to catch it.
Northflank's Vibe-Coding Playbook Has Four Stages. bex Only Ships Two of Them by Default
Northflank's enterprise vibe-coding guidance reduces to four jobs a platform owes an AI-generated app: audit, harden, observe, sandbox. Here's which two bex ships as defaults today, and which two are still the tenant's job — checked against the actual bex.yml schema.
SPIFFE/SPIRE for AI Agents: Cryptographic Workload Identity Instead of Long-Lived Service Account Tokens
A hands-on look at issuing an AI deploy-agent a cryptographic identity distinct from a human's with SPIFFE/SPIRE — and the honest operational cost next to the Vault/External-Secrets-Operator setup most self-hosted platforms already run.