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 26-30 of 235 articles.

Human-in-the-loop

HITL and the Reciprocity Problem: Why Reviewer Decisions Bias Toward the Customer Even When Policy Suggests Otherwise

Jul 18, 2026Human-in-the-loop

HITL and the Reciprocity Problem: Why Reviewer Decisions Bias Toward the Customer Even When Policy Suggests Otherwise

Reviewers show customers mercy. The policy says reject, the reviewer approves with a note. The policy says escalate, the reviewer reassures. Reviewers don't apply policy neutrally — they apply policy with an unconscious tilt toward the customer's side. The reciprocity problem is the most pervasive reviewer bias in HITL, and the most invisible. Here is how it manifests, why it matters, and how to design for it.

Engineering

MCP Spotlight: Notion MCP Server — The Official Knowledge-Base Bridge With Markdown-Native Retrieval, Two-Tool Minimalism, and the Workspace-Context Default for Agents

Jul 17, 2026Engineering

MCP Spotlight: Notion MCP Server — The Official Knowledge-Base Bridge With Markdown-Native Retrieval, Two-Tool Minimalism, and the Workspace-Context Default for Agents

The official Notion MCP Server by Notion — 2 tools in the minimal profile (notion-search, notion-fetch), Markdown-native retrieval (12x token savings vs raw JSON), OAuth 2.0 with principle-of-least-privilege scopes, full CRUD available in the standard profile. MIT-licensed. The knowledge-base default for AI agents in 2026.

Product

Facio's Idempotency Discipline: How AI Agents Avoid Double-Charging Customers When Retries Happen

Jul 17, 2026Product

Facio's Idempotency Discipline: How AI Agents Avoid Double-Charging Customers When Retries Happen

AI agents fail and retry. The model returns 500, the agent retries the tool call. The network drops mid-flight, the agent retries. The tool returns a timeout but the operation actually succeeded, the agent retries. Without idempotency, retries cause duplicate side effects: double-charged customers, duplicate emails, duplicate records, double-deployed code. Facio's idempotency discipline gives every side-effecting operation a unique key, makes operations safe to retry, and detects duplicate executions before they cause damage.

Security

MCP Server Authentication in 2026: Why OAuth 2.1 with PKCE and Token Exchange Is Now the Only Compliant Pattern

Jul 17, 2026Security

MCP Server Authentication in 2026: Why OAuth 2.1 with PKCE and Token Exchange Is Now the Only Compliant Pattern

The MCP authorization spec mandates OAuth 2.1 with PKCE, RFC 9728 Protected Resource Metadata, RFC 8707 Resource Indicators, and RFC 8693 Token Exchange for zero-touch delegation. Token passthrough is the most dangerous anti-pattern. The compliant stack is the only path to production MCP servers in 2026.

Human-in-the-loop

HITL and the Latency Tax: Why Every Second of Review Waits Has a Cost the System Doesn't Acknowledge

Jul 17, 2026Human-in-the-loop

HITL and the Latency Tax: Why Every Second of Review Waits Has a Cost the System Doesn't Acknowledge

Every HITL review waits. The customer's request sits in the queue. The latency is real — the customer perceives it, the SLA measures it, the business pays for it. But most HITL systems treat the latency as external to the HITL design. The latency is internal. The latency has a cost. The cost is a tax on every decision the reviewer makes. Here is why the latency tax is the most underrated cost in HITL.