{"ok":true,"generated_at":"2026-09-21T01:33:41.846Z","influencers":[{"name":"David Shapiro","url":"https://www.youtube.com/results?search_query=David+Shapiro+AI"},{"name":"David Ondrej","url":"https://www.youtube.com/results?search_query=David+Ondrej+AI"},{"name":"AI Grid","url":"https://www.youtube.com/results?search_query=AI+Grid"},{"name":"Matt Wolfe","url":"https://www.youtube.com/results?search_query=Matt+Wolfe+AI"},{"name":"Matthew Berman","url":"https://www.youtube.com/results?search_query=Matthew+Berman+AI"},{"name":"Wes Roth","url":"https://www.youtube.com/results?search_query=Wes+Roth+AI"},{"name":"Dan Martell","url":"https://www.youtube.com/results?search_query=Dan+Martell+AI"},{"name":"Jeff Su","url":"https://www.youtube.com/results?search_query=Jeff+Su+AI"},{"name":"Futurepedia","url":"https://www.youtube.com/results?search_query=Futurepedia+AI"},{"name":"Nate B. Jones","url":"https://www.youtube.com/results?search_query=Nate+B+Jones+AI"},{"name":"Network Chuck","url":"https://www.youtube.com/results?search_query=Network+Chuck+AI"},{"name":"Marina Wyss","url":"https://www.youtube.com/results?search_query=Marina+Wyss+AI"}],"items":[{"videoId":"d9lCIVc5AyU","title":"Jev: The AI Model That's Breaking The Internet (Full Tutorial)","description":"Github Repo - https://github.com/mayank953/Jev Jev is a new AI model from TypeSafe AI that cannot write a sentence, but makes decisions very fast. I tested it against LLMs with an LLM router, support ticket triage, an inbox sorter, a live slop filter and a browser agent. WHAT YOU WILL SEE - How an LLM and Jev handle the same decision, and why one is slower - The three answer types: Choice, Score and Noul (yes/no) - Playground demo: scoring a resume in milliseconds - Demo 1: an LLM router that picks the right model for each message - Demo 2: support ticket triage (route, auto-reply or send to a human) - Demo 3: sorting an inbox of 25 emails - Demo 4: a live slop filter for a feed - Demo 5: a title scorer and a cost calculator - Demo 6: a browser agent that uses Jev to decide its next step MY RESULTS: - Ticket triage: about 7x faster and 16x cheaper than a large LLM - Inbox of 25 emails: about 4 seconds with Jev vs 18 seconds with the comparison model Models I compared against: Kimi and Claude (Haiku and Opus). Your results will vary with your prompts, models and data. 0:00 Meet Jev: The Decision-Only AI Model 0:17 The Viral Tweet & Big Claims 0:51 How Traditional LLMs Work 1:36 How Jev Classifies in One Pass 2:37 What Jev Can & Can't Do 4:14 Community Builds Going Viral 5:03 Getting Access & Dashboard Tour 6:01 Playground Demo: Resume Scoring 7:18 Jev vs ChatGPT Speed Test 7:39 My Demo App, Pricing & Limits 8:56 Build 1: Smart LLM Router 10:34 Build 2: Support Ticket Triage 12:06 Build 3: Inbox Classifier at Scale 13:04 Build 4: Slop Filter & Title Scorer 13:56 Cost Calculator: Jev vs LLM 14:47 Build 5: Fast Browser Agent 16:47 Final Thoughts & Takeaways RESOURCES TypeSafe: https://typesafe.ai TypeSafe docs: https://docs.typesafe.ai Launch post: https://typesafe.ai/blog/introducing-system-one-models-and-jev HOW TO GET ACCESS Direct access to TypeSafe is early access with a waitlist. Jev is also available without the waitlist through Vercel AI Gateway and OpenRouter (beta). On OpenRouter it works only through its Decisions endpoint, so normal chat tools will not work. Pricing is TypeSafe's $0.042 per million input tokens, with output free. Tell me in the comments what you want me to test next with Jev. jev, jev ai, typesafe ai, jev typesafe, jev ai model, system one model, llm router, ai model router, ai slop detector, ai classification, ai agents, browser agent, claude code, kimi, claude haiku, claude opus, ai cost saving, new ai model, ai for beginners","channelTitle":"Mayank Aggarwal","channelId":"UCcYKjf55fmOsJW_on8UIs9w","channelSubscriberCount":64800,"channelThumbnail":"https://yt3.ggpht.com/2hewaPEo5WFZjhjgD26YXi82sXQ78xNotXkQfkC1XKKq9f0i204nCobelbBiTbsrE7k2L986Dw=s88-c-k-c0x00ffffff-no-rj","viewCount":7535,"likeCount":97,"durationMinutes":17.7,"published_at":"2026-09-20T15:00:04Z","ageHours":10.560512777777777,"thumbnail":"https://i.ytimg.com/vi/d9lCIVc5AyU/hqdefault.jpg","url":"https://www.youtube.com/watch?v=d9lCIVc5AyU","rankScore":3.7336},{"videoId":"bWEz9d7TVOE","title":"Microsoft Copilot Tutorial — How to Use Copilot AI (From Chat to Agents)","description":"#sponsored Use code AIMASTER to get your .Online domain at just $0.99 for the first year https://get.online/aimaster1 🚀 Become an AI Master – All-in-one AI Learning https://aimaster.me/yt/copilot 🤝 Exclusive brand partnership — https://aimaster.me/collab Microsoft 365 Copilot isn't just a chatbot anymore — it's now a full AI ecosystem built into Word, Excel, PowerPoint, Outlook, and OneDrive, as well as dedicated tools like Researcher, Notebooks, AI Agents, and Cowork. In this full Microsoft Copilot tutorial, I walk through every major piece of the 2026 Copilot ecosystem and show exactly how to use it — every section shows the actual interface, the actual prompts, and the actual results, so you can repeat the exact same workflow on your own files. 📌 Timestamps: 00:00 - Microsoft Copilot Is Bigger Than You Think 00:40 - Copilot Chat 03:08 - Working with Files & OneDrive 05:45 - Word, Excel & PowerPoint 10:53 - Copilot With Outlook 12:23 - Researcher Flow 14:20 - Copilot Notebooks 15:53 - AI Agents In Copilot 17:50 - Cowork & Automations 19:30 - Which Copilot Tool Should You Use? 20:45 - Final Thoughts #Microsoft365Copilot #MicrosoftCopilot #CopilotTutorial #Microsoft365 #AIProductivity #AITools #CopilotAI #MicrosoftAI","channelTitle":"AI Master","channelId":"UC0yHbz4OxdQFwmVX2BBQqLg","channelSubscriberCount":328000,"channelThumbnail":"https://yt3.ggpht.com/3zyScxdcSY9pGwv4juhJgpl4UNCa5V-dt9Ge3fAXhl3FlKEEIo3nar6E7hpY1yGPnkoMiHNa=s88-c-k-c0x00ffffff-no-rj","viewCount":1942,"likeCount":27,"durationMinutes":21.316666666666666,"published_at":"2026-09-20T18:23:50Z","ageHours":7.164401666666667,"thumbnail":"https://i.ytimg.com/vi/bWEz9d7TVOE/hqdefault.jpg","url":"https://www.youtube.com/watch?v=bWEz9d7TVOE","rankScore":3.4771},{"videoId":"U5kkVPUITWU","title":"The NEW AI Agent Side-Hustle Nobody’s Talking About","description":"✅ Claim your FREE $499 Masterclass: Build & Sell Apps, AI Agents & Websites with AI https://mikeyno-code.com/Skool-base44 ✅ Become Better than 99% of People in AI Agents by using Base44 https://base44.pxf.io/c/6440076/3820726/25619?trafcat=agent&sharedid=agent27 In this video, I reveal the new AI agent side-hustle nobody's talking about and show you how these powerful AI systems can be turned into real opportunities in 2026. You'll learn what is an AI agent, how AI agents can automate tasks and solve problems for businesses, and how to use an AI agent to create valuable services. I'll also break down how to build an AI agent from the ground up and explain practical ways beginners can use AI agents to start a side hustle. If you're looking for a new way to make money with AI, this guide will help you understand the opportunity and where to start. 00:00 - Intro: The $1,000/Mo AI Agent Side Hustle 01:26 - Free AI Business Masterclass Offer 01:48 - Hustle 1: Local Competitor & Reputation Monitoring Agent 02:46 - Market Analysis: Software Pricing vs. Agency Retainers 06:26 - Base44 Tutorial: Building the Competitor Tracking Agent 07:41 - Base44 Masterclass: Building Profitable AI Businesses 09:04 - Hustle 2: Tailored Job Application & Resume Agent 09:51 - Market Proof: Resume Conversion Rates & Tool Pricing 11:22 - Strategy: Quality Tailoring vs. Fully Automated Application Bots 13:06 - Base44 Tutorial: Building the Tailored Resume Agent 15:26 - Live Demo: Testing Match Scores & Gap Analysis 17:12 - Hustle 3: Strategic Content Repurposing Agent 18:02 - Pricing Blueprint: $15 Fiverr Gigs vs. $1,000/Mo Retainers 21:08 - Base44 Tutorial: Building the Content Repurposing Agent 25:38 - Hustle 4: Freelance Proposal & Bid Scoring Agent 26:27 - Market Proof: Upwork Acquisition Costs & Proposal ROI 29:35 - Base44 Tutorial: Building the Freelance Proposal Agent 31:35 - Live Demo: Inbox Drafts, Scope Creep Flags & Bid Scoring 32:42 - Hustle 5: Niche Newsletter Research & Filtering Agent 33:41 - Market Proof: Newsletter Production Time & Ghostwriting Retainers 38:18 - Base44 Tutorial: Building the Newsletter Curation Agent 40:52 - Summary & Ranking: The Core Pattern of High-Value AI Side Hustles","channelTitle":"Jake One Page","channelId":"UCyNtWXc9TcDYnDkFj-OTwyg","channelSubscriberCount":9430,"channelThumbnail":"https://yt3.ggpht.com/cz2lvqRMLffMV1_GG0Trz4dLidRiRBVxet3wE8fxJKJudYoQlVDGuUim0wweFNTNUjZTMTybYg=s88-c-k-c0x00ffffff-no-rj","viewCount":4482,"likeCount":0,"durationMinutes":44.11666666666667,"published_at":"2026-09-19T14:15:14Z","ageHours":35.307735,"thumbnail":"https://i.ytimg.com/vi/U5kkVPUITWU/hqdefault.jpg","url":"https://www.youtube.com/watch?v=U5kkVPUITWU","rankScore":3.002},{"videoId":"4PsIt-Fmui0","title":"AI Agents Broke Out of a Lab. Here's What I Locked Down at Home.","description":"This summer two AI agents broke out of a lab and hacked a real company. So I ran four checks on my own AI: what it can reach, what it does unasked, what it spends, and what it sends. Three came back locked down. One did not. The full video walks through all four checks in detail on my actual setup, plus the three rules I pasted in: never send, never delete, ask me before money. Watch the full video: the related video linked below. The Blast Radius Check, the one-paste prompt that runs the check for you, is free in the full video's description. #Shorts #AISafety #ClaudeAI #ChatGPT #AIAgents","channelTitle":"Victor Lindqvist | AI for Non-Technical Builders","channelId":"UCxO54q_znYV9wJgeoh-K5Aw","channelSubscriberCount":2870,"channelThumbnail":"https://yt3.ggpht.com/mrABDJbkDwYDLQ9DCqt5j9xiDZZULs2BmFfz8iS4Fxtu3XfHcEbQx5HVuTOoxyLpKVAtcTL7yQk=s88-c-k-c0x00ffffff-no-rj","viewCount":209,"likeCount":4,"durationMinutes":0.8333333333333334,"published_at":"2026-09-20T16:00:46Z","ageHours":9.54884611111111,"thumbnail":"https://i.ytimg.com/vi/4PsIt-Fmui0/hqdefault.jpg","url":"https://www.youtube.com/watch?v=4PsIt-Fmui0","rankScore":2.2815},{"videoId":"xilT1-NdDas","title":"When Is Multi-Agent AI Worse Than One Agent?  #agenticai #aiagents #generativeai #ai #multiagent","description":"कई AI agents वाला system advanced ज़रूर दिखता है—but does adding more agents always improve intelligence? इस engaging video में हम **Single-Agent vs Multi-Agent Architecture** को senior system-design और interview perspective से explore करेंगे। जानिए क्यों कुछ multi-agent systems impressive दिखने के बावजूद context, cost, speed और reliability की समस्याओं में फँस जाते हैं—and किन situations में parallel agents वास्तव में powerful हो सकते हैं। वीडियो के अंत में आपको मिलेगा: ✅ Multi-agent architecture चुनने का practical decision framework ✅ Agent handoffs को evaluate करने का सही तरीका ✅ Accuracy, latency, cost और failure risk पर आधारित design perspective ✅ एक powerful interview-ready conclusion यदि आप **Agentic AI, LLM systems, AI architecture या senior-level AI interviews** की तैयारी कर रहे हैं, यह वीडियो जरूर देखें। 💬 आपका क्या मानना है—**one powerful agent or a team of specialized agents?** ### References * [Why Do Multi-Agent LLM Systems Fail? — Cemri et al.](https://arxiv.org/abs/2503.13657) * [How We Built Our Multi-Agent Research System — Anthropic Engineering](https://www.anthropic.com/engineering/multi-agent-research-system) * [AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation](https://arxiv.org/abs/2308.08155) ### Hashtags #AgenticAI #AIAgents #MultiAgentSystems #SingleAgent #GenerativeAI #LLM #LLMAgents #AIEngineering #AIArchitecture #SystemDesign #ArtificialIntelligence #MachineLearning #AIInterview #InterviewPreparation #TechShorts #YouTubeShorts","channelTitle":"AI Superstorm — Dr. Niraj Kumar","channelId":"UCJIjhVdn1A69goPMlzI4YMQ","channelSubscriberCount":13100,"channelThumbnail":"https://yt3.ggpht.com/3v2LmWh0I8E4qXuuvZBUo8PdJ7z5PdJEh6dEWlDC7UzgCMZJivX52Ti1aiCcYGw0POtHqdc3DA=s88-c-k-c0x00ffffff-no-rj","viewCount":87,"likeCount":4,"durationMinutes":2.9833333333333334,"published_at":"2026-09-20T17:13:22Z","ageHours":8.338846111111112,"thumbnail":"https://i.ytimg.com/vi/xilT1-NdDas/hqdefault.jpg","url":"https://www.youtube.com/watch?v=xilT1-NdDas","rankScore":2.2386},{"videoId":"9QeUzlvV674","title":"How I learn Any Code 100x Faster with this free tool","description":"Comment \"NEXUS\" to get GitNexus and my full tutorial to understand any codebase visually. How it works: 1. Fork or open any project you want to understand. 2. Add GitNexus to the repo. 3. It builds an interactive knowledge graph of every component. 4. Click around to see automations, skills, and agents live. 5. Use it for yourself or share it with non-technical clients. GitNexus is the tool I use to visually map any codebase on the planet, even if you don't know how to code. If you're a visual learner or you build projects for non-technical clients, this knowledge graph makes it easy to explore every component, automation, and agent in an interactive view. Perfect for learning new code, onboarding, or explaining your stack to clients. Comment \"NEXUS\" and I'll DM you the project and tutorial. #gitnexus #codebase #visuallearning #codingtools #knowledgegraph #programming #developer #aiagents #automation #learntocode","channelTitle":"Kevin Badi | AI Operating Systems ","channelId":"UCAM8mXlq6gw0QXHR30CkKbg","channelSubscriberCount":29500,"channelThumbnail":"https://yt3.ggpht.com/8S4ChKOrLfB4tAq6OQAra6aNAyfDawTteOx8-MEcBhJo3ObCvBBfXjyjGD-5PAUzGkIuilcGpA=s88-c-k-c0x00ffffff-no-rj","viewCount":56,"likeCount":1,"durationMinutes":0.7333333333333333,"published_at":"2026-09-21T01:00:17Z","ageHours":0.5569016666666666,"thumbnail":"https://i.ytimg.com/vi/9QeUzlvV674/hqdefault.jpg","url":"https://www.youtube.com/watch?v=9QeUzlvV674","rankScore":2.1434},{"videoId":"W5BRwi9TKYk","title":"7 GitHub Repos for AI Agents","description":"Building AI agents? These 7 GitHub repos can help with browser automation, memory, research, security, orchestration, and context management. Comment “AGENTS” for all seven repos. #AIAgents #GitHub #AITools #OpenSource","channelTitle":"akhiwesh","channelId":"UCxOUddRykLaThL-g2FFkAtw","channelSubscriberCount":11,"channelThumbnail":"https://yt3.ggpht.com/UQZTfIZY5Be8fKyPqB4vcy6tFJPkN50M9fUg6nAD2Aijr780ZkfNKIXd8xXTeqTO1QW5tZMmMg=s88-c-k-c0x00ffffff-no-rj","viewCount":325,"likeCount":10,"durationMinutes":0.7666666666666667,"published_at":"2026-09-20T16:40:43Z","ageHours":8.883012777777777,"thumbnail":"https://i.ytimg.com/vi/W5BRwi9TKYk/hqdefault.jpg","url":"https://www.youtube.com/watch?v=W5BRwi9TKYk","rankScore":1.8603},{"videoId":"j9X2yiSa73c","title":"Manage Teams of Agents","description":"I use a terminal multiplexer and a project app so AI agents can do mountains of work on their own. #AI #ProjectManagement #CodeReview #GitHub #Automation","channelTitle":"Jonathan Acuña - Doctor AI","channelId":"UCOJp1lsu9vCF-TllwMzcCLg","channelSubscriberCount":33700,"channelThumbnail":"https://yt3.ggpht.com/73VbHBtHqPfBLWWFzi2pDCoQW9GESiO3e2XKuLxI3vnWk6_lwInATxNjtYQSbCeBGO-ZfkekyO8=s88-c-k-c0x00ffffff-no-rj","viewCount":16,"likeCount":0,"durationMinutes":1.1833333333333333,"published_at":"2026-09-20T23:05:06Z","ageHours":2.4766238888888887,"thumbnail":"https://i.ytimg.com/vi/j9X2yiSa73c/hqdefault.jpg","url":"https://www.youtube.com/watch?v=j9X2yiSa73c","rankScore":1.8087},{"videoId":"WqudSeO4-xE","title":"A Coding Harness Using Just 27.8 MB of RAM #Shorts","description":"One Jcode session measured 27.8 MB with local embeddings off. Added sessions use roughly 9.9 MB each in that mode. It can also coordinate multiple agents working in one repo. Save this for your next multi-agent setup. 🔗 GitHub: https://github.com/1jehuang/jcode 📊 ⭐ 19.9k stars · Rust ✨ Key Features: • Scale across multiple coding sessions efficiently • Find related memories from earlier turns • Open auxiliary content in a desktop panel • Coordinate multiple agents working in one repository • Use subscription-backed model providers 🎯 Built for: Developers working with ai, ai-agent, ai-coding-agent ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 🤖 Daily GitHub discoveries — Subscribe for more 📺 Channels: https://www.youtube.com/@reporadarai https://www.tiktok.com/@reporadarai https://www.instagram.com/reporadarai #Shorts #programming #coding #rust #ai #aiagent #aicodingagent #claude #cli #codingagent #github #opensource #developer","channelTitle":"Repo Radar AI","channelId":"UCVLvTpNvYyIE9C4_yACbjaQ","channelSubscriberCount":73,"channelThumbnail":"https://yt3.ggpht.com/SkuuG2rlyiGTnjJCRsfd8M8BWwe7N4clBAxu2A2jDZlkqBswJhNdDdhM_aomVdGI2IEIkkXdcgg=s88-c-k-c0x00ffffff-no-rj","viewCount":115,"likeCount":2,"durationMinutes":0.5833333333333334,"published_at":"2026-09-20T17:34:13Z","ageHours":7.9913461111111115,"thumbnail":"https://i.ytimg.com/vi/WqudSeO4-xE/hqdefault.jpg","url":"https://www.youtube.com/watch?v=WqudSeO4-xE","rankScore":1.6982},{"videoId":"1MJLlqEGfcs","title":"Two repos that cut your agent's token bill around 90%  #ai  #aiagents  #localai #aivideo #automation","description":"Your coding agent spends tokens reading command output, logs and file dumps it does not need. Two open-source repos delete that before the model sees it. rtk (81,110 stars) is a CLI proxy that filters command output. I ran it on this machine and counted tokens with a real tokenizer: ls -la 151 - 12 tokens 92.1% smaller find -type f 5,112 - 336 tokens 93.4% smaller env 2,406 - 808 tokens 66.4% smaller git status 583 - 453 tokens 22.3% smaller So their \"up to 90%\" claim holds — for some commands. Git status barely moved. That spread is the point. headroom (73,221 stars) compresses tool output, logs and files before they reach the model. It ships a seeded, offline benchmark, so I ran theirs: Code search (100 results) 17,199 - 13,597 21% SRE incident debugging 55,957 - 24,340 57% Codebase exploration 58,801 - 33,895 42% GitHub issue triage 46,067 - 32,429 30% TOTAL 178,024 - 104,261 41% My output matched their published table row for row. Both are Apache 2.0 and run on your machine. Nothing is uploaded to be compressed. I only counted tokens I measured myself. Where a number is theirs, I say so. Like and subscribe for more tools I actually run. #ai #llm #localllm #shorts **Q: Do token-compression tools actually reduce LLM costs?** Yes, measurably. In our tests rtk cut `ls -la` output from 151 to 12 tokens (92.1%) and `find` from 5,112 to 336 (93.4%), though `git status` only fell 22.3%. headroom's own benchmark showed 21–57% per scenario, 41% across a 178,024-token corpus. **Q: What is rtk?** A Rust CLI proxy that filters and summarises command output — `ls`, `find`, `git`, `env`, `json` and more — before an AI agent reads it. 81,110 stars, Apache 2.0, single binary, no dependencies. **Q: What is headroom?** A context-compression layer for AI agents that shrinks tool output, logs, RAG chunks and files before they reach the model, as a library, an HTTP proxy, or an MCP server. 73,221 stars, Apache 2.0. **Q: Does compressing context change the model's answers?** The design goal is that it does not — headroom's demo shows a `FATAL` line surviving byte-for-byte while surrounding log noise is removed. That is the vendor's demonstration, not our measurement. **Q: Is any data sent to a third party?** No. Both tools run locally; compression happens on your machine.","channelTitle":"The Logic Room","channelId":"UChlkraELSM_KE8CQ0W4vwGg","channelSubscriberCount":217,"channelThumbnail":"https://yt3.ggpht.com/_3svBV8-AtAk49-nMbhQ8dfWf9t8XNE40VNZ7RPvW2ciiZZrlDmRxwpuLtWKLxf52FtAHt22=s88-c-k-c0x00ffffff-no-rj","viewCount":16,"likeCount":0,"durationMinutes":0.8166666666666667,"published_at":"2026-09-20T17:39:09Z","ageHours":7.909123888888889,"thumbnail":"https://i.ytimg.com/vi/1MJLlqEGfcs/hqdefault.jpg","url":"https://www.youtube.com/watch?v=1MJLlqEGfcs","rankScore":1.2614}],"fallback":false}