My Three AI Market Surprises for 2026 – Part 1

Every year we get to laugh at predictions that didn't pan out. So I'm throwing my hat in the ring. Who knows – maybe one or two of these hot takes will be closer to reality than most people think today?

We'll find out in January 2027.

Prediction #1: Open Source Agents Go Mainstream

2026 won't be the year of Linux on the desktop – but it will be the year of open source agents running open source language models. For consumers and businesses alike.

Why?

Hardware keeps getting more powerful. AI capabilities increasingly run on-device. And let's be honest: we're all carrying around a ton of computing power that we rarely fully use.

Add to that projects like OpenCode, Ollama, and LM Studio that make it incredibly easy to run all kinds of models locally.

The real surprise? The quality of open source models.

GLM, Qwen, DeepSeek and many others. GLM-4.7 in particular has impressed me in our agents – both in quality and speed.

The fact that almost all of these models come from China is an interesting twist, but that's a topic for another post.

What does this actually change?

You can already run agents around the clock today. The news is the price.

I can use spare resources on my own hardware to have agents working for me day and night – without ongoing API costs.

Our agents already work 24/7 across different projects: development, QA, market analysis, security…

My prediction #1: Open source agents on local hardware will go mainstream.

What do you think – can open source projects keep up with commercial models and agents? Or will it remain a niche?