Tag
A new survey puts AI agent rollback rates at 75%, climbing to 81% at the most mature organisations. That's not proof governance fails — it's proof governance is the only thing catching the failure at all.
A Cursor agent found one unscoped API token and wiped a production database and its backups in nine seconds. The real failure was credential scoping, not the model.
Vendors advertise 200,000-token context windows. The number production agents can actually use reliably is closer to 130,000 — and closing that gap is a compression-architecture decision, not a bigger-window one.
AP2, ACP, UCP, and x402 all claim to solve 'agentic payments.' They solve different problems, and one already failed in its first real deployment for reasons that had nothing to do with the protocol.
AI browser agents read web pages and user instructions through the same channel, so a hidden instruction in a Reddit post or calendar invite can hijack the agent. Patching the prompt won’t fix that.
MCP's auth is optional, tool definitions are mutable, and session-scoped permissions create ambient authority. Three attack classes, real CVEs, documented incidents — here's what to lock down before you ship.
YC W26 graduated 199 companies in March 2026. The interesting signal is not that 60% are AI companies — it's the breakdown: agent infrastructure, vertical agents, and foundational research. Here's what it tells you.
Klarna's AI handled two-thirds of all customer chats and the efficiency metrics looked clean. CSAT on complex interactions told a different story. Here is the diagnostic and the hybrid model that holds up.
Chroma's 2025 research tested 18 frontier models and found every one degrades as context grows. This is what context rot means for production engineering decisions — and the specific patterns that address it.
MCP is becoming the standard interface for connecting AI agents to external tools. But most teams adopting it don't have a clear picture of what the protocol covers and what it deliberately leaves out.
AI agents consume compute in ways that don't map to user count — and Copilot's June 2026 billing shift is the clearest signal yet. Here's what the transition reveals about pricing for AI-powered products.
Most teams discover that their AI agent has been burning money the wrong way after the invoice arrives. Five operational controls prevent that — and most teams build them too late.
Practical guides, product updates, and compliance notes — straight to your inbox. No fluff.
Newsletter is opening soon. We'll switch this on once we've got our first issue ready.