Every feed screams that you're falling behind on AI. New model, new app, ten more to try today. That panic is manufactured — someone is selling it to you. You're not behind. You're overwhelmed. And the fix was never more tools, it's the right three.
Three months ago I had thirty tabs open and no system. Now I run my whole day on three tools that work as one stack: one I think with, one that remembers everything for me, and one that does the boring work while I sleep. Here's exactly what each one is, how I use it, and how they connect into a single workflow instead of thirty lonely tabs.
The stack at a glance
| Tool | Role | What it does |
|---|---|---|
| Claude + Gemini | Think | Make you understand what you don't know; plan before acting |
| NotebookLM | Remember | A private, cited knowledge base you query for grounded answers |
| n8n / Zapier / VPS | Run | Automations that do the repetitive work on a schedule, while you sleep |
1. The one you THINK with
This is the front of the whole system: Claude and Gemini. Not "ask a question and copy the answer" — that's the shallow way everyone uses AI. The real move is to make it teach you what you don't know yet, and to make it plan before it acts.
Two ways I use it every day:
- Understand: when I hit something I don't actually get, I make Claude or Gemini explain it until I do — in my words, at my level, with examples. I leave understanding it, not just holding an answer.
- Plan first: before it does anything big, I make it lay out the plan — the steps, the order, what could go wrong. I approve the plan, then it executes. Planning first is what separates a useful AI from a confident mess.
Why two models? They have different strengths and blind spots. Claude for deep reasoning and building, Gemini for fast research and a second brain that sees it differently. Same job, two angles.
2. The one that REMEMBERS
NotebookLM is great on its own. You load in many sources — YouTube videos, PDFs, websites, your own notes — and it answers ONLY from them, with citations. It can turn that pile into summaries, study guides, even an audio voiceover overview. On its own it's already the best way to query a stack of material you trust.
But here's the part most people never reach — my actual workflow:
I query NotebookLM from Claude Code, and then I make Claude act on what it pulled.
Say I want to build a new Claude skill but I haven't fully learned the topic. I don't need to. I do this:
- Load NotebookLM with sources about that topic — YouTube videos that explain skills, websites and docs that cover them.
- From Claude Code, query that notebook for the grounded, cited answer.
- Tell Claude: using what you just pulled, build me the most relevant skill for this.
So NotebookLM becomes the knowledge base, and Claude becomes the builder that turns that knowledge into a real, working thing. I get to ship something I hadn't fully learned myself, because the AI pulled the real knowledge and acted on it. Query for a purpose, then act — that's the whole point.
The connector that wires NotebookLM into Claude is a free skill, credit PleasePrompto — the same one from my first video. Full step-by-step is in the companion how-to here: "How to Connect NotebookLM to Claude." Repo: github.com/PleasePrompto/notebooklm-skill
3. The one that RUNS while you sleep
The third tool is the automation layer: the repetitive work — sorting, summarizing, scheduled jobs, posting, scraping — handed off once, then it runs on its own. You wake up to it already done. There are three real ways to do this, and they trade ease for power and cost.
| Zapier | n8n | VPS (self-hosted) | |
|---|---|---|---|
| What it is | Hosted no-code automation | Open-source automation | Your own server running n8n / scripts / agents |
| Ease | Easiest — click and connect | Medium — visual, more control | Hardest — you run the machine |
| Power | Limited to its integrations | Very high — custom code, any API | Total — anything you can code |
| Cost | Per-task, expensive fast | Free self-host / cheap cloud | Cheap flat monthly |
| Best for | Quick wins, non-technical | Power users, no server bill | Heavy, always-on, private workloads |
What I use is a VPS. It's close to n8n — same idea of chaining steps and running them automatically — but instead of paying per task or living inside one app's limits, I have my own always-on server. No task caps, full privacy, a flat cheap monthly cost, and the freedom to run anything I can code, including AI agents overnight.
The trade-off is honest: a VPS is the most work to set up and you maintain it yourself. So the rule of thumb:
- Start on Zapier if you're non-technical and want results today.
- Move to n8n when Zapier's per-task pricing or limits start to hurt.
- Go VPS when you want full power, privacy, and a flat cost — and you're comfortable running a server (or letting Claude help you run it).
Tools & links
- Claude: the one you think with; deep reasoning and building. claude.com
- Gemini: your second thinking model; fast research, a different angle. gemini.google.com
- NotebookLM: the one that remembers; your private cited knowledge base. notebooklm.google.com
- The connector skill: wires NotebookLM into Claude, free, credit PleasePrompto. github.com/PleasePrompto/notebooklm-skill
- n8n: open-source automation, self-host or cloud. n8n.io
- Zapier: easiest hosted no-code automation. zapier.com
- VPS: your own always-on server, e.g. Hetzner or DigitalOcean. What I use to run automations 24/7.
What most people get wrong
They collect tools instead of building a system. Ten AI apps that don't talk to each other is just ten more tabs — more overwhelm, not less. The power isn't in any single tool, it's in the handoff: you think in the first, it remembers and pulls grounded knowledge from the second, and it ships the repetitive half through the third. Three tools that hand work to each other beat thirty that sit in isolation.
One step to take today
Close every AI tab except one. Pick the single tool you actually think in and make it the front of your stack. Add the memory layer once that's a habit, and the automation layer once you have a repetitive task worth handing off. Three on purpose beats a hundred by accident.