I Tried 100+ AI Tools. These 6 Are the Best
What this video is
⚡ a 10-minute video, readable in 60 secondsThe creator describes a personal knowledge system holding his projects, skills, meeting transcripts, and other knowledge, which he says drives about 90% of his daily workflow and has let him stop paying for tools like Gamma, Poppy, and a Loom equivalent. He built it on Claude Code and Codex as agents that read local files and build from natural language, routed through OpenRouter for model access and fallback, with Hostinger VPS hosting n8n and Hermes/OpenClaw agents, GrokBot cloud bots for scheduled tasks, Glido for voice dictation, Kie.ai for pay-per-use image and video generation, and Trigger.dev for scheduled scripts. He frames this as a setup he actively pays for and uses weekly, not a one-off demo, organized under an 'AI OS' framework he calls The HerkLoop built on four Cs: Context, Connections, Capabilities, and Cadence. Beyond one screenshot of his Skool community homepage, no further on-screen file structure or configuration detail was given, so most of this is narration rather than shown config.
System does (00:50): Personal knowledge system holds the creator's projects, skills, meeting transcripts, and knowledge.
Key takeaways
- System does (00:56): This system drives about 90% of the creator's daily workflow.
- System does (00:59): It replaced paid tools like Gamma (slide decks) and Poppy (content help) because he can now build equivalents himself.
- System does (01:09): He built his own version of Loom rather than renting someone else's tool.
- System does (00:41): Claude Code and Codex are used to make slide decks, spreadsheets, Google Docs, brainstorm, act as a council, and run the business operating system.
+ 40 more takeaways
- Stack (00:25): Tool one is actually two tools, Claude Code and Codex.
- Stack (00:27): Claude Code and Codex are described as AI agents that work directly on a computer, reading local files, writing code, and building from natural language input.
- Stack (01:38): Codex is said to be better at computer/browser use, while Claude Code is a better thought partner for brainstorming.
- Stack (02:13): OpenRouter (tool two) gives access to over 500 AI models from more than 80 providers through one account.
- Stack (03:22): The AI OS framework is called 'The HerkLoop,' built around four Cs: Context, Connections, Capabilities, and Cadence.
- Stack (03:13): A free 'AI OS' template kit with guided setup and six ready-to-use skills is offered, compatible with Claude Code and Codex.
- Stack (03:40): Tool three is Hostinger VPS.
- Stack (04:04): The creator runs n8n, Hermes agents, and OpenClaw agents on the Hostinger VPS.
- Stack (04:16): Hermes agents include one trading stocks and one making Instagram reels/carousels 24/7, plus others testing different Claude Code setups.
- Stack (05:11): Tool four is GrokBot, which shares a cloud computer with a browser, file system, and terminal.
- Stack (06:29): Tool five is Glido, a voice dictation app; the creator discloses being a partner in Glido.
- Stack (07:41): Tool six is Kie.ai, pay-per-use access to image/video models including VO, Kling, C-dance, Nano Banana, and GPT Image 2.
- Stack (08:15): Tool seven is Trigger.dev, which hosts TypeScript code and runs it on a schedule or via triggers like a webhook, with no server to manage.
- Prompts: no verbatim on-screen prompts, system instructions, or config text were present in the extracted points; graph-view labels naming specific skills or MCP servers were also not provided (not stated).
- Wiring (01:30): Pointing something like Claude Code at SOPs and client docs lets a team ask it questions instead of asking a person directly.
- Note (01:24): He suggests a good starting point for businesses is a tool paid for monthly but only used for one feature, offered as his own workflow opinion.
- Wiring (02:27): Automations running in production for his business and clients mostly use OpenRouter as the AI model provider.
- Wiring (02:38): OpenRouter lets him set up fallback models so if one goes down or errors, the automation tries the next one instead of breaking.
- Wiring (02:45): A separate API key can be created per automation so analytics show exactly which automation is costing how much money.
- Wiring (02:52): A suggested pattern uses a cheap fast model (Haiku) to sort incoming leads and a smarter model (Opus) to write replies or reasoning, all from one dashboard and provider.
- Wiring (03:08): Everything runs in one dashboard, can be billed on one credit card, and switching models is just changing one line of config.
- Wiring (04:23): Hostinger was chosen because a server can be spun up with one click without handling scaling, compute, or commands, since he isn't from a technical/DevOps background.
- Wiring (05:23): GrokBot has connectors built into apps already in use, speeding up automation setup.
- Wiring (05:31): GrokBot routines can be scheduled easily within the app.
- Wiring (05:39): One GrokBot grabs transcripts after every Q&A call in his Plus community, posts them in the community, and puts them in the classroom for later.
- Wiring (05:49): Another GrokBot posts in both the free and Plus communities every time he drops a YouTube video.
- Wiring (06:04): A 'Chieff of staff' GrokBot personality keeps him updated on AI news with reminder pings.
- Wiring (06:52): Glido's command mode lets it take action instead of just dictating, such as searching the web, summarizing a YouTube video, doing deep research, and finding/opening files and apps.
- Wiring (07:11): Command mode lets users build custom MCP servers (plugins) on top of Glido, including kicking off threads into Claude Code or Codex, or syncing to Notion or a task app.
- Wiring (08:28): Trigger.dev is used for deterministic, predictable automations like processing documents or syncing data between tools, which don't need a full Claude Code agentic harness.
- Wiring (08:48): The workflow is to build a script with Claude Code and put it in Trigger.dev, telling Claude Code the use case so it builds, deploys, tests, and updates the automation.
- Cost (04:58): Hostinger is running a Black Friday sale with up to 82% off, from now until about halfway through December, with a 30-day money-back guarantee.
- Cost (07:46): Kie.ai is marketed as trusted by 120,000+ developers in 80+ countries, with 99.9% uptime, 24/7 support, and $0 monthly fee.
- Cost (07:57): Kie.ai is used for carousel visuals, thumbnails, and Instagram Reels because you only pay for what you generate, making it the cheapest option he found for generating lots of images/videos at once.
- Cost (08:07): Kie.ai suits product photos, ad creatives, and short social videos with no monthly subscription, just pay-per-usage.
- Cost (09:08): Trigger.dev's free plan includes $5 of credits per month, and paid tiers start at $10/month for 50 concurrent runs plus additional features.
- Limitation (01:17): It's not easy to build software scalable to hundreds of thousands of users, though risk is lower for personal or small-team use.
- Limitation (01:58): The point isn't which AI tool is best overall, but which is best for the task at hand.
- Limitation (06:13): Running multiple GrokBots is compared to having many OpenClaws in your pocket but with far less technical setup, since OpenClaw requires figuring out server side, connections, and hosting.
- Limitation (09:26): These seven tools are the ones he has consistently used for about a year and a half to two years, though the list changes over time and may not fit every business.
Shown on screen — grab and go
CODEWeatherApp Next.js component code
Index.tsx
"use client";
import { weather } from "@/lib/weather"
import { SkyPlate } from "./SkyPlate";
import { HourlyStrip } from "./HourlyStrip";
import { TenDayForecast } from "./TenDayForecast";
import { DetailWidgets } from "./DetailWidgets";
import { TickTemp } from "./TickTemp";
export function WeatherApp() {
return (
<main className="relative min-h-dvh w-full">
<SkyPlate [...]OCR'd from the code editor screen at 05:19; trailing lines cut off with a gap marked [...] shown at 5:19
PROMPTYouTube comment analysis agent goal
Goal: Go through the YouTube comments on my last 10 videos [...] - Find comment fetch tooling - Pull comments from last 10 videos - Analyze comments for common themes - Report the one most important theme
merged from two overlapping OCR reads of the same agent dashboard around 02:29; the goal line is cut off mid-sentence and marked [...] shown at 2:29
PROMPTWeekly report generation prompt
Hey, generate this week's report
transcribed from the Claude chat input box visible on screen at 01:35 shown at 1:35
How this brief was shaped: Personal AI Build / System Demo · confidence Low
The speaker walks through their own business AI stack (Claude Code, Codex, OpenRouter with fallback keys) explaining how each tool functions inside their personal system, and the OCR shows a knowledge graph of their own workflow labels like mcp-server-integration, memory-2-autodream, and scheduled-tasks rather than any single tutorial or comparison table.
The lens sets this brief's structure, never its facts — every claim is held to the same citation and fact-check standard.
Jump to a moment
Their links, sorted & clickable
🏛️ Communities & courses2
How I built an AI agency from 0 to over $1 million per yearMy FREE resources💼 Sponsorship & business1
Sponsorship InquiriesWas this brief useful?