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 1-5 of 73 articles in Human-in-the-loop.

Human-in-the-loop

HITL and the Recency Trap: Why Reviewers Who Optimized for Last Quarter's Failures Will Miss This Quarter's

Every HITL team tunes their reviewers based on the failures they've seen. The tuning works for the failures that already happened. The tuning fails for the failures that haven't happened yet. The recency trap — over-optimizing for the most recent failure mode — is HITL's most predictable self-inflicted wound. Here is why the trap works, how it produces invisible failures, and how to design systems that stay current without becoming reactive.

Aug 7, 2026Human-in-the-loop

Read article

Human-in-the-loop

HITL and the Approval Refusal Problem: Why Reviewers Who Never Reject Are Not Engaged — They're Coasting

Aug 6, 2026Human-in-the-loop

HITL and the Approval Refusal Problem: Why Reviewers Who Never Reject Are Not Engaged — They're Coasting

Every HITL team has a reviewer who approves everything. The metrics look great: low latency, high throughput, zero escalations. The reviewer is celebrated. The reviewer is, in fact, coasting. The approval refusal rate is the most reliable signal of reviewer engagement — and the one most teams refuse to track. Here is why the absence of rejections is a red flag, not a green light, and how to design HITL systems that detect coasting before it becomes institutional.

Human-in-the-loop

HITL and the Epistemic Asymmetry: Why the Reviewer Knows Less About the Agent's Reasoning Than the Agent Knows About the Reviewer's Decision

Aug 5, 2026Human-in-the-loop

HITL and the Epistemic Asymmetry: Why the Reviewer Knows Less About the Agent's Reasoning Than the Agent Knows About the Reviewer's Decision

The reviewer sees the agent's final proposal. The reviewer doesn't see the agent's reasoning chain, the agent's uncertainty distribution, or the agent's alternative considerations. The agent sees everything the reviewer does: the decision, the reasoning, the timestamp, the audit trail. The information flows one way. The epistemic asymmetry produces rubber stamps, false confidence, and reviews that can't defend themselves. Here is why the asymmetry is HITL's most fundamental design flaw — and how to fix it.

Human-in-the-loop

HITL and the Decision Latency Exposure: Why the Time Between the Agent's Proposal and the Reviewer's Approval Is Itself a Risk Variable

Aug 4, 2026Human-in-the-loop

HITL and the Decision Latency Exposure: Why the Time Between the Agent's Proposal and the Reviewer's Approval Is Itself a Risk Variable

Every second between the agent's proposal and the reviewer's approval is a second the world changes. The customer's context drifts. The customer's emotional state shifts. The competitor's offer arrives. The agent's proposal ages. The reviewer's decision is being made on stale data, but the reviewer doesn't know how stale. Here is why decision latency is HITL's most underrated exposure variable — and how to design systems that treat time as a first-class risk signal.

Human-in-the-loop

HITL and the Asymmetric Cost of False Positives vs False Negatives: Why the Reviewer's Mental Math Almost Always Gets It Wrong

Aug 3, 2026Human-in-the-loop

HITL and the Asymmetric Cost of False Positives vs False Negatives: Why the Reviewer's Mental Math Almost Always Gets It Wrong

Every HITL decision is a tradeoff between false positives (approving a bad action) and false negatives (rejecting a good action). Most reviewers default to optimizing for the false positive — rejecting the bad action is visible, accepting the bad action is catastrophic. The asymmetric worry produces over-rejection. Here is why the asymmetric mental math is the most common collapse in HITL, and how to design systems that give the math the right shape.