Where AI actually is in 2026

Three years after ChatGPT, the press cycle is still where the press cycle was: agents, AGI, autonomous everything. Walk into a Norwegian SMB and the picture is different. Calmer. More boring. More useful.

What's actually deployed:

A managing director runs Microsoft Copilot in Outlook to draft replies and summarise threads. She doesn't think of it as AI. It's just how email works now.

An accounting team uses Visma's invoice scanning, which rests on document-understanding models that would have been a research paper in 2021. Nobody calls them AI internally. They're just "the bilag tool."

A sales team drops meeting notes into a Copilot prompt that updates the CRM and drafts the follow-up email. Two minutes saved per meeting × twelve meetings a week × forty weeks. That's a working week per year per rep.

A handful of teams run n8n flows that route incoming questions, escalate when needed, and write structured updates to Slack. Boring infrastructure. Quiet wins.

What's not deployed, despite being everywhere in pitch decks:

Bespoke fine-tuned LLMs trained on a single company's data. Almost no SMB in Norway is doing this. The cost-benefit doesn't work, and Copilot reading from your SharePoint is good enough.

Truly autonomous agents handling multi-step business decisions end-to-end. Mostly demos. The reliable production version is "AI drafts, human ships."

The "AI strategy report" as a deliverable. Strategy reports about AI got replaced — sometime in the last six months — by working pilots that take less time and prove more.

The interesting part of 2026 isn't where AI is going. It's where it already is, embedded in tools you already pay for, doing modest jobs without ceremony. The companies that win this year aren't the ones running custom inference at scale. They're the ones who turned on Copilot for everyone in March and built two boring n8n flows by July.

That's the boring news. It's also the only news that matters if you run a business.