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.

Clear

Showing 31-35 of 57 articles in Human-in-the-loop.

Human-in-the-loop

Right-Sized HITL: The Action Classification Engine That Decides What Actually Needs a Human

Jun 21, 2026Human-in-the-loop

Right-Sized HITL: The Action Classification Engine That Decides What Actually Needs a Human

The HITL industry is converging on a wrong assumption: that every AI action should have a human in the loop. The right architecture isn't 100% HITL — it's right-sized HITL. Most actions should be autonomous. Some should be sampled. A few need synchronous review. The challenge is mapping the right oversight to the right action.

Human-in-the-loop

Who Is Liable When the Human Approves a Bad Action? The Accountability Chain in HITL

Jun 20, 2026Human-in-the-loop

Who Is Liable When the Human Approves a Bad Action? The Accountability Chain in HITL

When an AI agent takes a bad action and a human approved it, who is legally liable? The agent's vendor, the deployer, the reviewer, the manager — the answer is unclear and inconsistent across jurisdictions. Here's how the accountability chain works, where it breaks, and what the audit trail must capture to make the answer defensible.

Human-in-the-loop

Five HITL Scaling Inflection Points: The Architecture That Breaks at Every Order of Magnitude

Jun 19, 2026Human-in-the-loop

Five HITL Scaling Inflection Points: The Architecture That Breaks at Every Order of Magnitude

The HITL roadmap that worked for five agents breaks at fifteen. The patterns that worked for one team don't transfer to five. The approval gates that worked for one action type break at thirty. Here are the five scaling inflection points and how to design for the order you'll hit them.

Human-in-the-loop

The Reviewer Doesn't Know What the Agent Knows: Closing the Information Asymmetry in HITL

Jun 18, 2026Human-in-the-loop

The Reviewer Doesn't Know What the Agent Knows: Closing the Information Asymmetry in HITL

Most HITL designs assume the human knows what's being asked of them. But the reviewer's mental model — what the agent is doing, why, with what risk — is often incomplete or wrong. The agent knows things the reviewer doesn't. The reviewer guesses. The guess becomes the decision. How to fix the information asymmetry.

Human-in-the-loop

HITL for Code Generation Agents: Why 'Looks Good' Approvals Are Creating Production Incidents

Jun 17, 2026Human-in-the-loop

HITL for Code Generation Agents: Why 'Looks Good' Approvals Are Creating Production Incidents

Code generation agents are the fastest-growing HITL use case — and the worst-implemented. Reviewers approve diffs they don't fully understand, miss subtle security flaws, and let prompt injection land in production. The 10× velocity claim isn't real if 30% of approvals are rubber-stamps.