AI Agents Blog
Deep dives and practical articles about AI agents — agent frameworks, orchestration patterns, MCP integrations, and autonomous workflows. Curated from our main blog: every card links to the full article. 840 articles and growing.
ServiceNow AI Research Launches DrBench: A Game-Changing Benchmark for Enterprise Deep Research Agents
Discover DrBench, ServiceNow's new benchmark that tests AI agents on realistic enterprise research tasks, pushing boundaries in multi-hop reasoning, tool use, and proprietary data handling.
Meta's ARE-GAIA 2.0 Redefines AI Agent Benchmarks for Real-World Asynchronous Challenges
Discover how Meta's ARE-GAIA 2.0 pushes AI agent evaluation into asynchronous, event-driven scenarios, setting a gold standard for measuring true agentic capabilities in dynamic environments.
Implementing a Secure AI Agent in Python: Self-Auditing Guardrails, PII Redaction, and Controlled Tool Access
Discover a robust Python framework for building AI agents that prioritize security through self-auditing mechanisms, automatic PII redaction, and restricted tool usage. This guide provides complete code and practical steps to deploy safe, enterprise-ready agents.
Unlock the Top 5 Agentic AI Design Patterns: Must-Know Strategies for AI Engineers in 2025
Dive into the exciting world of agentic AI with the 5 hottest design patterns transforming how we build intelligent systems. Boost your engineering skills with practical examples and GitHub repos!
ROMA by Sentient AI: Open-Source Meta-Agent Framework Revolutionizing Hierarchical AI Agents for AGI
Sentient AI unveils ROMA, a powerful open-source framework for crafting AGI-level agents with smart hierarchical task execution. Dive into its features, setup, and real-world potential to supercharge your AI projects.
Demystifying Computer Use Agents: Evolving from Web Automation to Full OS Interaction – Technical Breakdown and Practical Guide
Computer use agents represent the next leap in AI autonomy, enabling models like Claude 3.5 Sonnet to control computers via screenshots and actions. This guide dissects their mechanics, from web origins to OS mastery, with code examples for real-world deployment.
Google Unveils Open-Source MCP Server for Google Ads API: Empowering LLMs with Seamless Access to Ad Data
Google has open-sourced an MCP server that connects LLMs directly to Google Ads API, debunking myths about siloed ad data and enabling natural language queries for campaigns, performance metrics, and more.
Agentic Context Engineering (ACE): Empowering LLMs to Self-Improve Through Dynamic Context Evolution, No Fine-Tuning Required
Discover Agentic Context Engineering (ACE), a groundbreaking approach that enables large language models to enhance their performance on complex tasks by iteratively refining contexts, bypassing costly fine-tuning entirely.
Stanford's AgentFlow: Pioneering In-the-Flow Reinforcement Learning for Advanced Modular Tool-Using AI Agents
Stanford researchers unveil AgentFlow, a groundbreaking reinforcement learning framework that trains modular AI agents to excel in tool usage by embedding RL directly into their execution flow for superior performance.
Google AI Unveils Gemini 2.5 Computer Use Preview: Empowering AI Agents to Master Browser and GUI Interactions
Google AI's latest Gemini 2.5 preview introduces 'computer use' mode, letting AI agents navigate browsers and apps like humans via screenshots and precise actions. Open-sourced for devs to build powerful agents.
Anthropic Unveils Petri: Open-Source Framework Revolutionizing AI Model Auditing with Autonomous Agents
Anthropic's Petri framework automates AI safety testing by deploying intelligent agents to probe model behaviors across countless scenarios, scaling beyond manual red-teaming limits.
Google DeepMind Unveils CodeMender: Revolutionary AI Agent Leveraging Gemini DeepThink for Automated Patching of Critical Software Vulnerabilities
Google DeepMind's CodeMender uses Gemini's DeepThink to autonomously fix real-world C/C++ vulnerabilities, achieving an 81.7% success rate—outpacing prior methods by over 20%. Open-sourced on GitHub for developers worldwide.
Crafting Smart Conversational ML Pipelines: LangChain Agents Meet XGBoost for Streamlined Data Science Automation
Discover how to build an intelligent, chat-based machine learning pipeline using LangChain agents and XGBoost. Automate data science workflows from exploration to deployment with natural language interactions.
OpenAI Unveils Agent Builder and AgentKit: The Visual-First Toolkit for Creating, Launching, and Testing AI Agents
OpenAI just dropped Agent Builder and AgentKit, a powerful duo that makes building AI agents accessible to everyone—from no-coders using drag-and-drop visuals to devs crafting complex systems. Dive in to see how this stack revolutionizes agent development!
Crafting a Smooth Human Handoff Feature for AI-Driven Insurance Agents Using Parlant and Streamlit
Discover how to build an intuitive interface that seamlessly transfers complex insurance queries from AI agents to human experts using Parlant and Streamlit. Perfect for enhancing customer support in real-world insurance scenarios.
Scaling Software AI Agents with Minimal Data: Agency-Focused Supervision Delivers Breakthrough Results Using Just 78 Examples
Discover how a novel Agency-Focused Supervision (AFS) method trains powerful software AI agents with only 78 examples, achieving 37.4% on SWE-bench Verified—outpacing models trained on thousands. Perfect for developers building efficient agentic systems.
How to Evaluate Voice Agents in 2025: Going Beyond ASR and WER for Task Success, Barge-in, Hallucinations, and Noise Resilience
Traditional metrics like ASR and WER fall short for modern voice agents. Discover comprehensive evaluation strategies focusing on real-world performance in latency, interruptions, hallucinations, and noisy environments.
Agentic Design Methodology: Building Reliable, Human-Like AI Agents with the Parlant Framework
Discover the Agentic Design Methodology powered by Parlant, an open-source framework that enables developers to create robust, human-like AI agents through structured goals, planning, execution, reflection, and tools integration.
Google's TUMIX: Busting Myths on Test-Time Scaling with Multi-Agent Magic and Tool Mastery!
Discover how Google's TUMIX shatters limits in AI inference scaling using multi-agent teams and smart tool mixtures—boosting math reasoning by 20%+ over rivals!
Microsoft Unveils Open-Source Agent Framework: Revolutionizing Multi-Agent AI Orchestration with SDK and Runtime
Microsoft has launched the Microsoft Agent Framework, an open-source toolkit that streamlines building and deploying complex multi-agent AI systems. Integrated with Semantic Kernel, it offers robust orchestration, memory management, and tool integration for scalable AI applications.
AWS Unveils Open-Source MCP Server for Bedrock AgentCore: Supercharging AI Agent Building
AWS just open-sourced an MCP server tailored for Bedrock AgentCore, making it easier than ever to connect AI agents to powerful tools and data sources without the usual headaches.
The Future of Coding: Evolve from Engineer to AI Conductor and Supercharge Your Workflow
Discover how AI coding agents are transforming developers into conductors who orchestrate intelligent tools. From beginners exploring Aider to pros mastering multi-agent systems, unlock the future of software engineering today!
Crafting High-Performance AI Agents with Claude: Your Ultimate Step-by-Step Blueprint
Unlock the power of autonomous AI agents using Claude! Dive into a complete guide with tools, planning, memory, and execution to tackle complex tasks effortlessly.
Building Intelligent AI Agents Using Vercel AI SDK: A Comprehensive Developer Guide
Discover how to create advanced AI agents with Vercel AI SDK, leveraging tool calling, structured outputs, and streaming for real-world applications like task automation and multi-agent systems.