Blog
Human-in-the-loop for AI agents
Practical engineering and governance writing for teams running AI agents in production. Human-in-the-loop for AI agents means a person authorizes an agent’s high-risk action at runtime — before it executes — which is a different thing from the data-labeling sense of the term (humans annotating training data). We write about that runtime layer: approval gates, build-vs-buy, and tamper-evident audit trails.
- Invoice fraudJuly 17, 2026 · 7 min read
The IBAN Changed: Why the #1 Invoice Fraud Vector Is Still Working
An email arrives from a known supplier asking to update their bank account. The invoice looks legitimate. The IBAN is different. Your ERP processes it because it trusts the document — not your master data. This is the most successful invoice fraud vector in existence, and it takes one comparison to stop.
Read article - Our pivotJuly 17, 2026 · 7 min read
From "Govern Any AI Agent" to "Every Invoice Checked": Why We Pivoted
The product did not change. The policy engine, the Slack integration, the hash-chained audit trail — all the same. What changed was the problem we pointed it at. Nobody searches for "AI governance." Everyone searches for "why did we pay this invoice twice?"
Read article - Invoice fraudJuly 17, 2026 · 8 min read
Approval Laundering: The Invoice Fraud Hiding in Plain Sight
Three invoices of €1,900 from the same supplier in 24 hours. Each one is under your €5,000 approval threshold. Each one looks legitimate. Together they are €5,700 — and nobody noticed. That is approval laundering.
Read article - LandscapeJune 28, 2026 · 12 min read
AI Agent Approval and Human-in-the-Loop Tools: The 2026 Landscape
The approaches to human-in-the-loop for AI agents fall into a handful of categories, and they are not really competing for the same job. Here is an honest map of the 2026 landscape and a way to choose the right one for your stack.
Read article - FrameworksJune 28, 2026 · 13 min read
Human-in-the-Loop Across Agent Frameworks: LangGraph, CrewAI, Vercel AI SDK, and More
Every major agent framework now ships a way to pause for a human. None of them ships the durable routing, the policy engine, or the tamper-evident audit you need around that pause. Here is what each gives you, and what you still have to wire yourself.
Read article - EU AI ActJune 28, 2026 · 14 min read
Human Oversight of AI Agents Under the EU AI Act (Articles 12, 14, 26)
If your AI agent is high-risk, the EU AI Act requires human oversight you can demonstrate and logs you can produce. Here is what Articles 12, 14, and 26 mean in practice — and how to satisfy the mechanisms without slowing your agents down.
Read article - Audit & complianceJune 28, 2026 · 13 min read
How to Make Your AI Agent Audit Trail Tamper-Evident
A logs table is not an audit trail. Here is how to build a record of agent decisions that an auditor will accept — append-only, hash-chained, and verifiable by anyone — plus an honest read of what the EU AI Act Article 12 actually demands.
Read article - Build vs buyJune 28, 2026 · 12 min read
Building Human-in-the-Loop Approval for AI Agents Yourself: The Hidden Costs
The hard part of human-in-the-loop for AI agents is not pausing the agent — it is everything around the pause. Here is what you are actually signing up to build and maintain, and the point where buying becomes the rational call.
Read article