Facio Blog

Practical notes on human-reviewed AI agents.

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 36-40 of 281 articles.

Human-in-the-loop

HITL and the Triage Question: Why the First Five Seconds of Review Determine the Rest More Than the Next Five Minutes

Jul 28, 2026Human-in-the-loop

HITL and the Triage Question: Why the First Five Seconds of Review Determine the Rest More Than the Next Five Minutes

The first five seconds of a review determine the rest more than the next five minutes. The reviewer's initial read, the pattern match, the gut reaction — these set the trajectory. Everything after is justification. Here is why the triage question matters, what the first five seconds contain, and how to design HITL systems that respect the reviewer's intuition without abandoning the policy.

Engineering

MCP Registry Landscape 2026: From `mcp://` URLs to Verified Catalogs, and Why Curated Distribution Beats npm-Universe Trust

Jul 27, 2026Engineering

MCP Registry Landscape 2026: From `mcp://` URLs to Verified Catalogs, and Why Curated Distribution Beats npm-Universe Trust

The MCP registry landscape in 2026 — Docker MCP Catalog (production default, signed + provenanced), Official MCP Servers (Anthropic-maintained references), Glama (discovery), Smithery (community + npm), mcp.so/AIPOLABS (curated lists). Defense-in-depth distribution with signature + provenance + isolation + RBAC + HITL.

Product

Facio's Backpressure Discipline: How AI Agents Adjust to Capacity Without Drowning the System

Jul 27, 2026Product

Facio's Backpressure Discipline: How AI Agents Adjust to Capacity Without Drowning the System

AI agents produce work. The downstream — databases, APIs, humans, dependent agents — has capacity limits. The naive approach floods downstream regardless, causing cascading errors, latency spikes, and human review burnout. Facio's backpressure discipline gives agents mechanisms to sense downstream capacity and adjust production rate: real-time capacity sensing, adaptive throttling, queue management with overflow strategies, flow control signals, and backpressure-aware retries. The agent flows at the pace the system can absorb.

Human-in-the-loop

HITL and the Disagreement Problem: Why Two Reviewers Seeing the Same Action Often Means At Least One Is Wrong

Jul 27, 2026Human-in-the-loop

HITL and the Disagreement Problem: Why Two Reviewers Seeing the Same Action Often Means At Least One Is Wrong

Two reviewers see the same action. One approves, the other rejects. The disagreement is in the audit trail. The team has to decide: who is right? Most teams pick the senior reviewer and hope. The right approach is to treat the disagreement as a signal, not a problem. The disagreement reveals where the policy is ambiguous, where the context is insufficient, where the reviewer's calibration is off. Here is why disagreement is HITL's most informative event — and how to design for it.

Engineering

MCP Spotlight: Cloudflare MCP Server — The 2,500-Endpoint Edge Bridge With Code Mode, Two-Tool Minimalism, and the API-Default Reference for Agents

Jul 26, 2026Engineering

MCP Spotlight: Cloudflare MCP Server — The 2,500-Endpoint Edge Bridge With Code Mode, Two-Tool Minimalism, and the API-Default Reference for Agents

The official Cloudflare MCP Servers by Cloudflare — 2 tools in the Code Mode profile (mcp_search, mcp_execute) giving access to all 2,500+ Cloudflare API endpoints with bounded context. Plus 13+ specialized product servers. MIT-licensed. The edge-infrastructure default for AI agents in 2026.