Blog

Latest

{ / Discover our insight/ ]

Stay up-to-date with our latest blog posts.

Human-in-the-loop AI: the approval that sits between a proposal and a live action

Most human-in-the-loop AI puts a person after the model and before the reader. In an agentic system the loop belongs somewhere else entirely: between the proposed action and the executor that would run it.

What a policy engine decides, when a human is actually required, and why an audit log is not a chat log.

Governance
Trust

Human-in-the-loop AI: the approval that sits between a proposal and a live action

What a policy engine decides, when a human is actually required, and why an audit log is not a chat log.

Governance
Trust

Read the articleRead more

{ / all blog posts / ]

How AI agents complete real actions — safely.
Policy checks, approvals, and an audit trail on every step.

Autonomous AI agents: autonomy is a permission model, not a personality

An agent that only answers is not autonomous — it is fast search with a friendly tone. Autonomy starts where the agent is allowed to do something in a real system.

The autonomy spectrum, why agents need action boundaries rather than process rules, and the three ways autonomy reliably breaks.

Category
Agents

7 min read

AI agent guardrails: govern what the agent does, not just what it says

The guardrails market is mostly about language: hallucination filters, prompt-injection defences, moderation, model risk. Useful work — and completely silent on whether an agent may issue a €480 refund or open a production database.

This is the other half: guardrails on actions. What the agent is allowed to do, who signs it off, and what is written down afterwards.

Governance
Actions

8 min read

A managed alternative to self-hosted AI — and an honest look at what you give up

If you are searching for a self-hosted AI agent or an on-prem LLM, you are almost certainly not shopping for a server. You are shopping for control: your data staying yours, your systems staying yours, and no vendor holding you hostage.

We do not offer self-hosting — and we would rather explain that clearly than sell around it. Here is what the requirement underneath usually is, and who we are honestly not for.

Deployment
Data residency

7 min read

An AI agent for customer service that resolves, instead of deflecting

Support teams are still judged on deflection — how many tickets never reached a human. Customers do not want an answer; they want the refund issued, the plan changed, the access granted.

A walk-through of one duplicate-charge ticket, resolved end to end on a trust chain: the agent proposes the action, a policy engine checks it, a human approves it, an executor runs it through your billing API, and every step lands in an audit log.

Support
Execution

8 min read

Enterprise RAG: knowledge that completes the action

Almost anyone can build document search with a well-cited answer on top. Retrieval has become the easy half of enterprise RAG.

The value starts where retrieved knowledge turns into a completed operation in a live system: a grounded proposal, checked by a policy engine, executed through your API, and written to an audit log.

Capability
Knowledge

7 min read