Agentic AI

Ever been woken up by a 3 AM pager alert? We walk through how to build a multi-agent AI system with Haystack that can autonomously detect, investigate, and even write postmortems for production incidents.

AI agents are powerful, but they can be incredibly expensive. Let's walk through how to build a smarter agent that balances quality with real-world constraints like token limits, latency, and API costs.

Ever seen an AI confidently misread a tiny detail in a photo? Google's new Agentic Vision in Gemini 3 Flash aims to fix that, turning AI from a passive observer into an active detective that can use tools to get the right answer.

The Allen Institute for AI (AI2) just dropped SERA, a new family of open-source coding agents that compete with massive closed systems. We break down their clever and surprisingly efficient training method that could change the game for repository-level AI.

AI agents need a reliable way to communicate. Let's ditch the complex, stateful sessions and build a clean, secure, and asynchronous protocol that lets our agents work together, scale effortlessly, and handle long-running tasks without breaking a sweat.

Ever wish your AI assistant could actually do things, like clean up your downloads folder? Anthropic's new Cowork feature for Claude does just that, giving the AI a sandboxed space to work directly on your local files.

Ever wish your AI assistant could do more than just send you links? Google's new Universal Commerce Protocol (UCP) is an open standard that gives AI a "shared language" to complete purchases for you, from checkout to delivery tracking, all inside a chat.

AI agents often struggle with memory, forgetting key details or getting lost in context. New research from Alibaba and Wuhan University introduces AgeMem, a framework that teaches LLMs to manage their own memory.

A new open-source project called SETA is making waves by providing a complete environment for training and testing AI agents that live in the command line. Let's break down why this is a bigger deal than it sounds.

OpenAI is asking freelance contractors to upload projects from their past jobs. It's a bold strategy to teach AI how to handle real office work, but it comes with a major catch you need to know about.

We've all been obsessed with bigger, more powerful AI models. But what if that's the wrong race? A new AI software engineer from Meta and Harvard shows that the system *around* the model is what truly matters, and the results are pretty stunning.

Forget human training. A new kind of AI is learning by getting curious and asking its own questions. This could be a massive leap forward, but what does it really mean for us?