The pace of AI releases in 2026 is genuinely hard to keep up with. A model drops, everyone has takes, a week later there's a new one. If you try to evaluate everything, you evaluate nothing. The filtering matters more than the following.
My actual system: I follow a short list of people who do primary research rather than commentary. A handful of engineers who work with these tools seriously and write about what they find. The ratio of takes to signal in AI content is worse than any other technical domain I've followed. The rest comes from using the tools daily — you develop a feel for capability faster that way than through reading about it.
Some of the sources I read: Latent Space for technical depth, Ethan Mollick for research-backed practical takes, Simon Willison for hands-on tool exploration, AI Report for a Dutch perspective, and the Anthropic Economic Index for data on how AI is actually being used at scale.
