Where to read AI agents news

In 2024, most of us were still arguing about whether chatbots could write a decent email, and just two years later, my grandma is publishing iOS apps with the help of AI agents.

Things move so fast that keeping up feels like a red queen's race. You have to run faster and faster just to stay in place. "Omg, the new OpenAI model is the best! Ah no, now it's Claude now! Wait, Google just announced a new Gemini thing!" The trick is knowing where the signal lives, because the AI news machine thrives on hype, smoke, and mirrors.

This guide covers some of the channels worth following (as of 2016) if you want to keep up with AI agents news without becoming the sort of person who says "stochastic zero-shot" to feel smart.

2026 is the year agents got real

AI agents are LLMs armed with tools. A slight improvement over monkeys with typewriters. They do this looping thing where they think, use a tool, think again, use a tool again, etc, until they solve whatever you ask them to do. Why do they need tools? You might ask. Well to interact with the outer world, of course. They've been trained with static data, but you want them to access data they've never seen before, or even modify the state of the world. Something as simple as knowing what time it is. An LLM can't know that, but with a tool they can. A bit like us, isn't it? If you are not into calculating shadow angles, a watch is a pretty useful tool to have around.

So yeah, to summarize agents are like normal chatbots, but they can browse websites, write code, send emails, fill out forms, extract data, and work through tasks that need more than one step. That distinction matters because companies are now deploying agents for jobs with boring names, which is how you know the technology has become serious.

Why is this happening in 2026? Well, AI labs have understood that the killer app of AI are not chatbots but agents, so they've teached new models to become better at long-horizon tasks and tool calling. That's mostly it. Models are not really that more intelligent, they just got more agent-like skills.

Developers saw it first. At the start of the year, Claude Code become really popular and now most developers are not writting code by hand anymore. The same pattern is spreading beyond code, that's why Anthropic and OpenAI are taking their developer products and re-packaging them for b2b use cases: excels, pdfs, emails and all that office work stuff.

Computer use pushed the story further. Agents can now see screens and click buttons working around the lack of APIs of some vendors, or the unclear authorization policies for agents.

The sources worth following

You do not need to follow everything. Just pick the few sources that ressonate with your particular interests and discard the rest.

Non-technical sources

These sources explain AI and agents without assuming you are a machine learning researcher nor a developer.

Newsletters

• The Rundown AI is a daily digest of major AI news. It is useful when you want the broad map without spelunking through every announcement yourself.
• Superhuman focuses on practical AI tools and workflows. Good for seeing what people can actually use today, not just what a lab claims is about to eat the universe.
• The Neuron mixes news with business context. It is readable, fast, and usually good at separating useful updates from theatrical fog.
• Import AI, from Jack Clark, is more reflective. It covers policy, industry moves, research, and the longer arc of the field. Bring coffee. Not because it is dull, but because your brain will be asked to sit upright.
• Ben bites is a good source for AI news and insights for developers.
• Alpha signal brief daily AI news and updates, for people without time to read the whole internet.

Podcasts

• Hard Fork from The New York Times covers AI alongside the rest of tech culture. It is a good choice if you want context without needing to live inside the machine learning discourse bunker.
• The AI Daily Brief is short and frequent, usually the right size for a commute, a walk, or pretending you are "staying current" while waiting for a deploy.
• Practical AI focuses on real applications. It is less interested in mystical pronouncements and more interested in what teams can build and ship.
• AI for Humans is a fun weekly podcast about AI for non-technical people. It is a good choice if you want to understand AI in a way that is easy to understand.
• The most interesting AI podcast in the world perhaps the most important podcast of all time.

YouTube

• Matt Wolfe reviews AI tools and explains major changes in a way non-technical viewers can follow.
• The AI Advantage is strong on tutorials and productivity use cases. Useful if you want to see the tools in motion.
• Sabrina Ramonov a good source for AI news and insights for entrepreneurs.

Technical sources

These assume you are comfortable with engineering and implementation details.

Newsletters

• The Batch from Andrew Ng covers research and industry news with enough technical detail to be useful without becoming punishment.
• Davis Summarizes Papers does what the name says. It saves you from opening a PDF at midnight and whispering "just the abstract" like a liar.
• Ahead of AI from Sebastian Raschka (author of the book "Build a Large Language Model (From Scratch)") is excellent for machine learning practitioners who want research explained with care.

Podcasts

• Latent Space is one of the better places for deep conversations with builders, researchers, and engineers working close to the metal.
• Gradient Dissent from Weights & Biases features practitioners talking about the real implementation problems.
• Machine Learning Street Talk goes long. Very long. Good for technical depth if you like interviews that take their shoes off and stay a while.
• Dwarkesh Patel probably some of the best interviews with the most interesting people in the field. Some episodes are accessible to a non-technical audience, but others go deep.

YouTube

• Yannic Kilcher is useful for paper reviews and technical explanations. Dense, but worth it when a paper matters.
• Two Minute Papers gives visual research summaries. It is a good way to understand what a new system does before deciding whether to read further.
• Andrej Karpathy the legend. He is not only one of the greatest AI researchers of all time, but also an excellent teacher.
• Fireship covers tech news quickly, with enough humor to keep the weekly AI churn from feeling like a compliance training module.
• Stat Quest is a great channel for learning about statistics and machine learning. Bam!

Where the real conversations happen

Curated sources are useful, but the best texture often comes from communities where people show what broke.

• Hacker News is still the default place where technical founders and engineers gather to be skeptical in public. AI agent threads often produce better comments than articles. Sometimes the comments are insufferable. Sometimes they save you three weeks. Such is civilization.
• Reddit is noisy, but there are plenty of subreddits worth following.
• Discord servers for specific AI tools are often where practical truth leaks out first. Users share workflows, report failures, test features, and complain in a way marketing pages never quite manage.
• GitHub trending shows what developers are actually building. Watch which agent frameworks, browser automation tools, and orchestration libraries gain stars quickly. Stars are not truth, but sudden attention usually means there is something worth inspecting.

How to decide what matters

Most AI news will not matter to you. This is emotionally difficult because every headline now needs to "change everything" or won't survive the attention wars. Most things are not that important, and change nothing. Just wait a little bit and you will quickly find out which ones survive the test of time, and which ones were just hype.

I hope this guide helps you!