I want to be upfront about something before you read this post.
My AI setup runs on Claude Code and Obsidian. Everything. Writing, research, client work, newsletter drafts, my entire repo of ideas. It works really well for me, and based on my performance data, Claude Code posts have been some of the best-performing content I’ve published. A real segment of this newsletter already lives in that world with me, and I’ll keep writing for that group.
But I also hear from the other half of you all the time. You don’t want to open a terminal. You don’t want to spend weekends writing custom commands and sub-agents. You want AI that fits the tools you already use, without a second job as a configurator.
For a long time I didn’t have a clean recommendation for that group. “Just use ChatGPT” felt too shallow. “Build your own Claude Code setup” felt too deep. The middle path was missing.
That’s why I had Anfernee from Solopreneur Code on One Shot Show last Wednesday. Dheeraj Sharma and I wanted to see someone who’d solved the same problem I solved, but from a completely different direction. Anfernee runs his entire business inside Notion AI. No Claude Code. No terminal. No custom scripts. He relies purely on Notion with one AI agent he calls Nova, and years of context.
Solopreneur Code is Anfernee’s twice-weekly newsletter for solopreneurs who want to earn more and work less. Every issue covers practical systems, AI workflows, content creation, and digital products. Subscribe free and grab the Solopreneur Success Hub — ready-to-use systems that save 20+ hours a week.
If you want to go deeper after reading this post, start with these three:
Watching his screen made me realize that the real problem I’ve been solving isn’t a technical one. It’s something I think a lot of you are quietly suffering from too, no matter which tools you use.
The copy-paste tax nobody talks about
You’ve probably heard some version of this advice: “Use the best tool for the job. Claude for writing. ChatGPT for reasoning. Gemini for Google stuff. Perplexity for research.”
It’s a half-truth. Yes, different models have different strengths. But every time you cross a tool boundary, you pay a tax. You manually carry the context over. And that tax compounds.
Think about what you do in a single writing session:
You look up your own past notes (Notion, Obsidian, or wherever your thinking lives)
You research the topic (Perplexity, Google, ChatGPT)
You draft (Claude, ChatGPT, or your writing tool)
You save the output (back to Notion, Google Docs, Substack)
If those four places are four different apps with four different AI models, you’re doing the work of the AI. You’re the one remembering what the research said. You’re the one re-pasting your voice guidelines every session. You’re the one holding the thread together.
You’ve become the middleman in your own business.
If this is the future promised by AI, I don’t think it will change how we work.
The reframe I needed
Here’s what Anfernee said on the show that reframed this for me:
“To me, context is a lot more important than whether the LLM name is powerful or not.”
That one line is the whole thing. Every AI company is racing to have the smartest model. Every Twitter feed is arguing about Opus vs. GPT-5 vs. Gemini 3 Pro. Meanwhile, the builder who’s 99% committed to one place, with years of history already loaded in, is running circles around people jumping between fresh chat windows.
The question stopped being “which AI is best?” It became “where does my work actually live, and can I let the AI come to me instead of me going to it?”
Two answers to that question work. Mine is Claude Code plus Obsidian. Anfernee’s is Notion AI. They look different on the surface, but they solve the same underlying problem: one home, AI that lives in it with you, no middleman work.
Since my path takes a long detour through terminal-land, the rest of this post is about Anfernee’s. But before we get into how he built Nova, a quick primer for anyone who hasn’t looked at Notion AI recently.
First, What is Notion AI?
Notion AI is the AI layer built directly into Notion. It lives in the sidebar, reads every page you have access to, and does the things you’d expect: summarize, draft, answer questions, generate content. What makes it different from opening Claude or ChatGPT in a separate tab is that it already knows everything in your Notion. No copy-paste. No file uploads. No “here’s the context you need first.”
Three things matter about where Notion AI is today:
Custom instructions. You can write a personality and rules for how the AI should behave (the equivalent of a CLAUDE.md file if you use Claude Code). Whatever you put in the instruction file is applied to every chat. This is the feature that turned Notion AI from “meh” into “actually worth using” when it launched late last year. Anfernee’s Nova is built on top of this.
Custom agents. Released in early 2026. You can now spin up specialized agents with their own instructions, triggers, and scheduled runs. Think sub-agents, but no-code. Anfernee’s “modes” (content writer, decision maker, strategy) are the pattern that predates Notion’s official custom agent release, and they still work the same way.
Model choice. Notion AI lets you pick the underlying model: Claude Opus, GPT-5, Gemini 3 Pro. On certain plans, Opus is unlimited inside Notion AI. That alone is a big deal if you’ve ever hit the Claude Pro usage cap at 10 a.m. and had to wait four hours.
Notion AI is the fastest way to have AI that actually knows your work, without building anything technical. If your notes, meetings, and projects already live in Notion, you’re already 80% of the way there. You just haven’t turned it on properly yet.
That’s what the rest of this post is about.
🔒 What’s inside this AI Maker Lab post
This is a paid-tier deep dive because it unlocks the full One Shot Show replay plus the custom instruction templates Anfernee shared live.
Here’s everything you get below:
🎥 The full 59-minute live replay (paid subscribers only)
Watch the entire One Shot Show episode with Anfernee. Timestamps included so you can jump straight to any moment: the Nova walkthrough, the COMPASS framework, the live weekly reflection demo, the Notion AI vs Claude Code Q&A.
📄 Anfernee’s Nova custom instruction template
The full personality, memory, and mode structure he spent three days writing. Grab it, adapt it, paste it into your Notion AI settings. The same file runs his business today.
📄 The weekly reflection prompt
The one-line instruction Anfernee types every Sunday that replaces half a day of manual weekly review. Includes the exact structure it generates (activity map, patterns, scorecard, next-week plan) so you can run it on your own Notion.
⚙️ The COMPASS framework
Anfernee’s self-iteration protocol for long multi-step tasks (his record: 50 tasks in one instruction). Why it works the same way Claude Code plan files work, and how to build your own version.
🧠 The decision framework: Notion AI vs Claude Code
The honest breakdown of which path fits which kind of builder, based on how your current tools actually map to one home or the other.
⚠️ The trade-offs I won’t sugarcoat
Where Notion AI is slow. Where it breaks. Why Dheeraj moved off Notion MCP to the Notion API. Why I left Notion entirely for Obsidian. And why, despite all that, Anfernee’s path is still the one I’d recommend to most of you.
If you’re tired of being the router between six AI tools, this is the post that fixes it.
Meet Nova
Anfernee shared his screen and walked us through Nova.
Nova is his custom Notion AI agent. The name (Neural Operative Virtual Assistant) was suggested by Notion AI itself when he asked it what it wanted to be called. He also asked it to design its own logo. He’s been building Nova for years inside his Notion, which by his count holds roughly 15 million blocks across two years of writing, client work, meeting notes, and ideas.
Here’s what’s inside Nova’s custom instruction file:
