Skip to content
Insights and strategy

The Usermode blog

Practical thinking on autonomous AI, operations and building a business that runs itself.

Featured
Autonomous operations8 min read

Copilots wait to be asked. Your operations can't.

Most "AI for business" is a smarter chatbot that helps a person who is still doing the work. This essay argues that for operations, waiting to be asked is the wrong default — and lays out what it takes to hand a workflow to an autonomous fleet that runs it to completion on a schedule, with a human approving.

T
The Usermode team
Usermode
Read article
Trust & governance7 min read

Enterprise AI security without the badges

An honest account of Usermode's enterprise security posture today — per-tenant Azure isolation, an OAuth 2.1 gateway, signed external-send authorisations, prompt-injection sandboxing and redacted logs — and why we lean on inherited Azure certifications plus zero data custody rather than a badge wall.

28 Apr 2026
Buying AI10 min read

What is agentic AI? A buyer's guide for mid-market teams

Agentic AI is software that plans and completes multi-step work toward a goal — on a schedule, under approval gates, with an audit trail. Vendors rarely lead with the buyer's half: realistic 2026 cost bands, where agents pay back first in a 50–500-person firm, and the ten due-diligence questions that expose a rebadged chatbot. This guide covers exactly that, from the buyer's side of the table.

10 Jul 2026
Self-hosted AI10 min read

Open-weight AI models in 2026: a business buyer's guide

Frontier-class AI models you can download and run on your own hardware now sit roughly three to six months behind the closed labs — and two of the best ship under MIT or Apache 2.0. This guide maps what leads the open-weight field in 2026, which licences you can actually deploy under, what hardware runs what, and gives a straight answer on the China question — organised by business job, not benchmark score.

1 Jul 2026
Buying AI10 min read

What is MCP? The Model Context Protocol explained for operators

MCP is the open standard — created by Anthropic in November 2024 and since adopted across the industry — that lets AI models plug into business systems through reusable connectors instead of bespoke integration projects. Everyone repeats the USB-C analogy; what matters mid-evaluation is what happens after the plug fits: the governance layer deciding what the model's new hands may touch. Here is MCP in plain English, the real security surface, and the eight questions to ask any vendor whose deck says they support it.

27 Jun 2026
Buying AI11 min read

AI agents vs RPA: the difference, and when each wins

The standard answer to “agents or RPA?” comes from RPA vendors, and it is always “both — on our platform”. Here is the vendor-neutral decision rule instead: rules-stable, structured work stays on RPA; judgement, unstructured input and heavy exceptions go to an agent. Plus the two running costs nobody quotes — the brittleness tax and the variance tax — and how governing each actually differs.

30 Jun 2026
Trust & governance11 min read

GDPR and data sovereignty for AI agents: the 2026 guide

The EU AI Act's high-risk obligations apply from 2 August 2026, and most of what is written about the deadline is either a law-firm summary or a sovereignty pitch. This guide maps each obligation — GDPR minimisation and transfers, Article 12 logging, the CLOUD Act's reach — to the concrete technical control that satisfies it, and is honest about when the answer is hardware you own.

6 Jul 2026
Trust & governance11 min read

How to test AI agents: evals, golden traces and CI gates

AI agents don't fail like normal software — they fail politely, with plausible output and the wrong action, which is why the demo tells you almost nothing. This guide translates eval engineering into management language you already trust: golden traces as the probation exam, code scorers and LLM-as-judge as the marking scheme, and CI gates as no keys until you pass. It ends with the five eval questions to put to any AI vendor.

29 Jun 2026
Trust & governance11 min read

Human-in-the-loop approvals that don't kill speed

Put a human in front of every agent action and within a month you have either a bottleneck or a rubber stamp — the documented failure mode is a manager waving through 200-plus requests a day at a near-100% approval rate. The fix is to treat approvals as a designed product with a latency budget: risk-tiered gates, evidence packs that let an approver decide in seconds, signed single-use authorisations, and the approval rate itself watched as a health metric.

7 Jul 2026
Get started

Put your operations on autopilot

See your own fleet running in a 30-minute demo. We'll map the roles to your business and show you exactly what they'd do on day one.

Book a demo