|
agentic ai cringelords
Eric Morrison
The video is a satirical critique by Eric Morrison mocking the trend of social media influencers who post viral videos showing off elaborate AI "agent command centers" (dubbed "Jarvis" or "Hazel") that supposedly run their entire businesses autonomously. Morrison highlights how these setups—featuring spoken daily reports, swarms of sub-agents, and dozens of dashboards—appear staged, pointing out that one creator's agent delivered the exact same report word-for-word in multiple videos. He argues that while AI agents are genuinely useful technology, a culture of hype and one-upmanship has emerged where grifters exaggerate capabilities to push a fear-of-missing-out narrative, leading real people to buy into the false idea that they must automate everything or be "left behind."
|
|
What Is Agentic AI? | Agentic AI Tutorial For Beginners | Agentic AI Explained | Simplilearn
Simplilearn
This Simplilearn tutorial explains agentic AI through the story of Emma, a small business owner who discovers the difference between standard AI chatbots and agentic AI. While a normal AI only provides advice (e.g., "run discounts and post on social media"), agentic AI goes further by understanding a goal, analyzing relevant data, creating a multi-step plan, and producing actionable outputs like email drafts, social media captions, and campaign calendars. The video emphasizes that agentic AI operates across many domains—marketing, studying, software development—but stresses that human oversight and approval guardrails are essential to prevent the AI from taking high-stakes actions autonomously. The tutorial concludes by promoting Simplilearn's agentic AI certification as a way to build structured skills in working with these systems.
|
|
Pydantic AI 2.0: The New Best Way to Build AI Agents is Composing Capabilities
Cole Medin
Pydantic AI 2.0 introduces a central new primitive called the **capability**, which bundles an agent's instructions, tools, lifecycle hooks, guardrails, and model settings into a single composable unit. The idea is that AI agents can now be constructed by combining multiple capabilities like Lego blocks, making it easy to reuse components across different agents and projects. The presenter argues this concept supersedes MCP servers (which are now a subset of a capability) and positions Pydantic AI ahead of competitors like LangChain, Crew AI, and coding-agent SDKs such as Claude's. A GitHub repository with side-by-side Pydantic AI 1.0 and 2.0 examples is provided as a practical starting template.
|
|
Anthropic Quietly Dropped a Playbook to Secure Your AI Agents (Zero Trust)
The AI Automators
Anthropic published a 36-page zero-trust security playbook for AI agents like Claude Code, addressing the growing risk that autonomous agents—given broad access to data, tools, and outbound actions—create major attack surfaces. The video explains a "lethal trifecta" concept: when an agent simultaneously accesses private data, reads untrusted content, and can take outbound actions, attackers can exploit this combination through techniques like prompt injection; a Cloud Security Alliance study found 98% of 100 production agents were exposed to all three conditions, with only 11% passing a basic security bar. The presenter breaks down why traditional perimeter ("castle") security fails for agents—they operate at machine speed, read untrusted text constantly, and hold standing credentials—arguing that a zero-trust approach ("never trust, always verify") is necessary, with controls that completely remove capabilities rather than merely adding friction. The video also notes some cynicism about Anthropic's motives, since the same company releasing powerful models that expose cybersecurity vulnerabilities is now promoting its own security scanner as the solution.
|
|
LLM Wikis and how to give your agents memory
LangChain
This LangChain webinar brings together three guests — Brace (LangChain), Devon Stein (CEO of DOSU), and Jeff (Chroma) — to discuss LLM wikis and agent memory. Brace demos **OpenWiki**, a new npm CLI tool that automatically generates and maintains personal or codebase memory by ingesting data from sources like Notion, Gmail, and Slack via a nightly cron job. Devon demos **DOSU**, a knowledge infrastructure platform that plugs into coding agent workflows via MCP, capturing learnings from agent tasks into concise, human-readable "knowledge index" pages that are automatically injected into future agent sessions and kept up to date as code changes. Jeff frames the problem from Chroma's perspective, arguing that wikis represent a new kind of index — layering a "road network" of connections on top of unstructured data to enable agents to traverse knowledge more efficiently than raw vector embeddings alone allow.
|
|
Agentic AI Frameworks Explained: Workflows, Multi-Agent, & Production
IBM Technology
This IBM Technology video explains that choosing the right agentic AI framework depends on the type of system you want to build, not on which framework is universally "best." It outlines five categories of agentic AI systems: linear workflows (suited for LangChain, LlamaIndex), autonomous multi-agent systems where agents collaborate on open-ended goals (AutoGen, CrewAI), role-based systems where agents have clearly defined responsibilities (CrewAI, AutoGen), production orchestration systems requiring deep API and business integration (Semantic Kernel, LangGraph), and rapid prototyping using visual, drag-and-drop tools (LangFlow, Flowise). The core advice is to first identify what kind of system you're building—predictable pipeline, collaborative team, or production deployment—and then select the framework optimized for that use case.
|
|
KEEP AI LOCAL! Explaining Agentic AI and The Loop
Level1Techs
The video explains agentic AI and "closing the loop" — a setup where an AI model autonomously controls another computer to accomplish a defined goal with specified success criteria. The host demonstrates a local implementation using an NVIDIA DGX Spark (running the Neotron model) to remotely control an MSI mini PC, with agent frameworks like Hermes and Turnstone acting as orchestrators that coordinate multiple AI instances simultaneously. Turnstone, a project the host has contributed to, is highlighted for its enterprise-focused safety features, including a judge model (Gemma from Google) that monitors AI actions, a permissions/risk system, and "skills" files that teach the AI how to use specific tools or scripts. The host cautions that while agentic AI is exciting, it remains a hype cycle with real limitations — AI has skills but lacks judgment — and human oversight and durable, guardrailed processes are essential rather than simply turning AI loose.
|
|
Kittl's Agentic AI Just Changed The Game 🤯 (Print on Demand Tutorial)
Adrian Vonarx
The video is a tutorial by Adrian Vonarx demonstrating Kittl's new "Agentic AI" feature for print-on-demand sellers on platforms like Etsy, Shopify, and Amazon. Agentic AI simplifies t-shirt design creation by automatically selecting the best AI model, writing detailed prompts, and generating finished designs from just a few plain-English words, saving users time and AI tokens compared to manual prompting. The tutorial covers three methods: the "quick and easy way" (typing a simple phrase like "t-shirt design for female book lovers"), the "strategic way" (uploading a screenshot of a bestselling Etsy design as inspiration for an original niche design), and a third "variants way" for generating multiple design options. Adrian also compares Agentic AI favorably against standard Kittl models like Nano Banana Pro, showing that Agentic AI produces more polished, print-ready results from the same minimal prompt.
|
|
I Love the Karpathy LLM Wiki but it Doesn't Scale. Here's What Does.
Cole Medin
The video argues that while markdown-based "second brain" personal agents (like Karpathy's LLM Wiki approach) are popular and useful for individuals, they fundamentally cannot scale to production environments where multiple users need access, because they lack governance, access control, auditability, and cost efficiency. For production agents shipped to real users, the creator advocates replacing the markdown file system with a proper database architecture built around two core components: a **context retriever** (giving the agent structured access to business data) and **agent memory** (short- and long-term memory per user). As a concrete demonstration, the video uses Redis and its new "Redis Iris" tooling—including an MCP server for context retrieval and built-in agent memory—layered under a Pydantic AI agent built on mock e-commerce data, with the creator noting that Pydantic AI remains his preferred framework for production agents over coding-focused SDKs like Claude Code due to speed and token efficiency.
|
|
Local AI Coding Agents Are Finally Good Enough
Code with Beto
The video demonstrates that local AI coding agents have become genuinely useful with the release of Qwen 3.6, a 27-billion parameter open-source model optimized for agentic coding that supports vision, tool calling, and reasoning while requiring lower VRAM than previous models. The presenter shows how to set it up using LM Studio on a Mac and connect it to the OpenCode coding agent, then runs live demos including scanning a real iOS app codebase, adding new screens, building a Minesweeper and Snake game, and identifying SF Symbols from screenshots. While the model requires significant hardware (the presenter uses an Apple M3 Ultra with 96GB RAM and notes it was too demanding for an 18GB MacBook Pro), it performs comparably to Claude 4.5 on agentic benchmarks and works entirely offline. He also highlights LM Studio's LM Link feature, which lets users chat with their local model remotely from a phone.
|