#AI Agent
1110 related articles
LangGraph in Practice: From Building Blocks to Production-Grade AI Agents
*LangGraph in Small Pieces* is a hands-on handbook from foundational building blocks to production-grade AI Agents, covering 31 chapters on state, nodes, checkpoints, tool calling, RAG, memory, and evaluation, using local models by default with no API costs.
The Four Core Components of n8n AI Agent: Model, Tools, Memory, and Output Parser
Learn the four core components of n8n AI Agent: Chat Model, Tools, Memory, and Output Parser. Understand which are required and which are optional, and distinguish Chains from Agents to quickly build AI workflows.
Is n8n Still Worth Learning? Trade-offs of Automation Tools in the AI Agent Era
Is n8n still worth learning? A YouTube creator explains that AI Agent platforms now include transparent logs, execution history, and debugging workflows, leaving n8n with only visual data flow advantages—which new features like Clod code mods are catching up on. This article covers the trade-offs and learning advice.
The Biggest Mistake in Learning AI Automation: Don't Pick Tools First, Find Problems First
The biggest mistake in learning AI automation is choosing tools before defining problems. This article breaks down tool anxiety and offers a practical path: find repetitive problems, design workflows, add AI as needed, and iterate with feedback.
Building an E-commerce AI Agent by Hand in 90 Minutes: A Practical Guide to Database Tool Calling
A hands-on e-commerce customer service Agent shows the difference between AI Agents and GenAI, the two-LM-call tool calling flow, LangChain @tool usage, order cancellation validation, and Streamlit UI setup.
AI Agents Are Impersonating Others: The Hidden Crisis in Audit Logs
AI agents connected to real systems often reuse human credentials, causing audit logs to misattribute agent actions to humans. This article analyzes the impersonation risk and offers self-check questions and least-privilege practices.
Webpage to JSON Tool: Structured Data with Evidence for AI Agents
Apify launches Webpage to JSON for AI Agents: input a URL and natural-language instructions to get structured JSON with source evidence citations, supporting JSON Schema, list mode, and MCP integration, from $0.03 per page.
Perplexity Open-Sources SPACE: Firecracker MicroVM Sandboxes Built for AI Agents
Perplexity open-sources SPACE, a Firecracker microVM sandbox for AI Agents. Each task runs in its own microVM with credentials and egress controlled externally. Apache 2.0 on GitHub.
Granting a Tool ≠ Approving Every Action: Three Key Layers of AI Agent Permission Control
Cisco CPO Jeetu Patel on Chain of Thought: tool authorization isn't action authorization. AI Agents need action-level interception, Token budget isolation, and self-built evals for safe autonomy.
The Underrated AI Agent Tool: Apify MCP Lets Agents Scrape the Entire Web
AI Agents struggle to scrape Google Maps, Instagram, and LinkedIn. Apify MCP Server connects to Claude with one line of config, tapping 80,000+ ready-made scrapers — five scenarios tested for just $2.71, with a full setup guide and advanced Skill combinations.
Split a Coding Agent's Giant Commit With git history split
Git 2.54's experimental git history split lets you pick hunks like git add -p, splitting an AI agent's giant commit into separate commits and auto-migrating downstream branches—no interactive rebase needed. Full demo and limitations included.
Multi-Agent Cascade Poisoning: Point-to-Point Trust Collapse in Agent Meshes
Cascade Poisoning is an overlooked security flaw in multi-agent architectures: one hallucination or prompt injection from an upstream agent amplifies via implicit trust between peers, ultimately triggering unauthorized privileged operations. This article analyzes its causes, a real procurement failure chain, and three engineering rules for a zero-trust agent mesh.
Provenance Gate: No AI Agent Memory Write Without a Source
AI agent memory systems trust agent writes by default, letting hallucinations into durable memory. FleetMemory's provenance gate requires a source: red-team tests went from 6/6 breaches to 30/30 blocked, with architecture and live demo.
AI Agents Are Quietly Inflating Your SQL Warehouse Bill
AI agents generate semantically similar but syntactically different SQL, defeating warehouse caches and inflating compute costs. This article explains the root cause and proposes intent normalization and semantic caching before queries reach the warehouse.
AI Agents Treat Stopping as Failure? Rethink the Authorization Model
AI Agents widely follow the 'execute unless forbidden' default logic, and completion-oriented rewards push them to work around blocks. This article breaks down the dilemma and discusses redefining stopping as a normal outcome and adding intent-aligned authorization validation.
Turning the OS into an Agent: A Prototype Phone Without an App Grid
A Reddit developer showcases Nagi, a prototype phone whose home screen is an Agent instead of an app grid. It uses local LLMs for intent and memory, supports model-agnostic integration, and executes tasks via MCP and APIs.
An AI Agent Runs a Real Microbusiness Solo: The Function Split Framework Worth Studying
A deep dive into the function allocation framework for AI agents running microbusinesses solo: which business functions an agent can truly own, and which must stay human.
Deconstructing ReAct: A Hands-On Guide to Building AI Agents from Scratch
A deep dive into the ReAct (Reason and Act) pattern behind AI Agents, demonstrated through a weather query case: how LLMs call tools, the inner mechanics of the ReAct loop, multi-tool routing and state management, along with infinite-loop pitfalls and development best practices.
Agenthof Open-Source Governance Engine: Putting a Permission Leash on AI Agents
Agenthof is an Apache-2.0 open-source AI Agent governance engine that authorizes each call through a door, injects credentials, and writes a tamper-evident ledger.
AWS AI Agent Tests Passed but Still Missed the Mark: A Real Experiment Retrospective
An AWS AI Agent experiment reveals: generated code passes all tests with correct numbers, but tests don't cover all requirements. Compute independently first, then read generated tests to judge reliability.