---
title: "Operator Notes on AI-Native Products"
url: https://nativefirst.ai/
description: "Teardowns, benchmarks and playbooks on building AI-native products. From an operator who runs everything on agents."
author: John Tan
publisher: nativefirst.ai
format: Markdown copy of the web page at the url above, same text.
---

# Everyone’s using AI. Only 1% are all in.

Notes on building AI-native products and agents.

[Read the Playbook](https://nativefirst.ai/blog/)

Daily stack

- Claude Opus 5.5
- Codex GPT-6
- Cursor
- Grok Bot
- Instinct
- Jev
- gstack
- Lovable
- Granola

## Latest from the Playbook.

- [Google DeepMind Verified Signal's Code in Four Weeks. It normally takes years.](https://nativefirst.ai/blog/alphaproof-code-signal/): AI Models · 5 min read
- [What Is ARC-AGI?](https://nativefirst.ai/blog/what-is-arc-agi/): AI Benchmarks · 5 min read
- [Fable 5.2 Built a Browser Shooter in a Few Hours.](https://nativefirst.ai/blog/fable-52-browser-shooter/): AI Models · 4 min read

[All 162 posts](https://nativefirst.ai/blog/)

## Companies are trying AI. Most are stuck at Level 1–2.

1. 01
   ### Thought partner
   One-off prompts.
2. 02
   ### Personal assistant
   Faster tasks, same org.
3. 03
   ### Agent teammates
   Work accelerates. Margins lift.
4. 04
   ### Agent operating system
   The org runs on agents.

Level 2 to 3 is crossable in weeks, **one function at a time.**

## Companies deployed AI. They bought subscriptions.

- **30%** ChatGPT & Copilot adoption
- **0** workflows changed
- **0** proven ROI
- **Monthly** committee, nothing shipped

Who I am

## I don’t write about agents. *I run on them.*

**John Tan** nativefirst.ai

Building

Founder, [wonderfulco.ai](https://wonderfulco.ai/)

Before

CEO, Depict (Y Combinator). The team went on to found Lovable and Legora

How I work

Runs all work and personal life on agents by default
