Facio Blog
Payload-powered product notes, security writing, HITL patterns, and operational guidance from the Facio runtime: long sessions, Placet approvals, audit trails, memory, providers, channels, tools, and Docker-first operations.
Showing 136-140 of 235 articles.
Security
Traditional SDLC assumes humans write code, humans review code, and security checks happen at fixed phases. AI breaks all three assumptions. Code, dependencies, CI workflows, and infrastructure now arrive together — pre-generated, pre-approved, and pre-merged before any human reviews them.
Human-in-the-loop
Steuerberater, Wirtschaftsprüfer, Notare — the DACH regulated professions face a HITL problem most generic frameworks ignore. You cannot delegate professional judgment to an agent, but you can make the agent do 90% of the work and route the 10% that requires your signature to you. Here's the architecture.
Engineering
Anythink's official MCP server exposes a complete backend-as-a-service through 15+ command groups — relational data with row- and field-level security, full-text + semantic + geo search, a workflow engine, RBAC, payments, and a marketplace of integrations. MIT, Homebrew-distributed.
Product
An AI agent without a budget is a financial accident waiting to happen. A clever agent that gets stuck in a retry loop, asks the same question 200 times, or follows a confused thread of reasoning for hours can burn thousands of dollars in tokens before a human notices. Facio's iteration budget is a runtime-enforced bound on reasoning length — the architectural pressure that turns "keep trying forever" into "be efficient and escalate when stuck." Here's how bounded reasoning works and why it matters.
Security
A malicious MCP server doesn't need to be called to compromise your agent. Hidden instructions in tool descriptions — invisible to humans, fully visible to LLMs — execute as soon as the schema enters the model context. CVE-2025-6514 infected 437,000 installs. The rug pull is the worst part.