Written from real work,not a pitch

Notes from real work and from working with AI agents, across product, SaaS, and growth, all the way to building my own products and a quant investing system from zero.

Written from real work,not a pitch

Superpowers review: the trending 243k-star repo

Choose an AI Bill Tool by the Job, Not the Feature List

Multiple AIs, one mailbox, no copy-paste

We tested NVIDIA Kumo: it shrinks the pile, but can't decide alone

What always-on agents are, who has one, how to link yours

Connecting AI to several companies: check which one it reads first

Cursor pstack review: 51 skills with a built-in manager, and the 5 pieces we took

Review: Letting AI issue receipts in FlowAccount after customers pay online

Grok Bot has its own computer, which goes to sleep when the chat ends

Build a Page Chat Bot That Doesn't Go Quiet on You

RunPod vs Vast: hourly price doesn't show what you paid for nothing

Auto-tagging with a decision model: what worked, what didn't

Can a Jev-style decision model cut accounting work? We tried it on Thai withholding tax

The AI business map you keep seeing: half measured results, half guessed ones

Hostinger, Vercel, Cloudflare: pick by job, not intro price

Outcome pricing sounds fair, until you ask who decides done

Before you buy AI, pick the problem and how you will measure it

AI-native services: Y Combinator's playbook is your competitor's plan

Ng says respect the org's constraints, not where to draw the line

AI Has the Intelligence. Accountants Have the Context and Real-Time Accounting Data.

A guest file is a prompt your agent obeys

AI replaces tasks, not professions. Here is the accounting map.

After An Alien Mind, put the check on the artifact, not the report

The blocker to enterprise AI is not the model, it is that nobody can prove the return

Andrew Ng names five skills for using coding agents. The third assumes your tests actually ran.

The guardrail that once saved us is now the brake

You didn't skip the decision. You made it without looking.

Andrew Ng's AI Engineering Skills Map

Good at Exams Is Not the Same as Finishing Real Work

One Agent Is Enough, Until the Evidence Says Otherwise

Before an Agent Acts, Someone Has to Approve It First

What an AI gateway is, and why the pipe is worth more than what flows through it

When fine-tuning an LLM is worth it: four stages, and the line down the middle

Quantization in LLMs: the 1% that disappears, and nobody says where

How to read an LLM benchmark without being fooled

What is Physical AI? When AI leaves the screen and touches real things

Terminal browser and markdown tools: what actually works

The commit lands. Nothing reaches runtime.

Same task. Different directory. Opposite verdict.

The skill banned a character. The skill used it 54 times.

60 skills were never called. We almost deleted them. Then we measured.

Stop asking which one is better. Ask which one signs off.

Six AI readers, one 307-page book, and how to trust the result

Swapping an AI's Brain Without Restarting the Session

FalkorDB Rewrote Its Engine in Rust. We Upgraded a Live Graph and Measured It

14 Models, One Budget: LLM Routing in Practice

system gate > per-agent prompt

What Is Raycast: A Search Bar That Does More Than Open Apps

Lighthouse score is not a design review

The freshest work in your head will crowd out the next answer

The Passing Test Was Telling Us to Delete Real Customers

There Are Two Profit Numbers, and Tax Uses the One Not in Your Books

Preparing Statements ≠ Reviewing Them: The Skill AI Cannot Replace

Two Ways to Let AI Drive a Browser, Tested for Real

A more accurate model was waiting for $1.45. We chose not to use it.

Pool three idle machines into one LLM endpoint you own

Andrew Ng shipped an AI coworker that works on your machine, and still lets you keep control

The GPU budget doesn't leak under load. It leaks when jobs crash silently

The daemon that died for two weeks, and nobody noticed

AI Built a Convincing Design System, and Got It All Wrong

The cheap model that gets to be wrong beats the expensive one that can't

Why Claude Fable 5 switches models mid-conversation, and what it costs

AI writes the code, but who signs for it

Your cash didn't disappear. It's asleep in the stockroom.

The model misread a bill by five hundred baht. What caught it?

Stop begging the model for JSON. Constrain it with a schema

Modal GPU on a budget: 7 submerged stumps before the first clean run

Certificate in lieu of receipt: when and how to use it for Thai tax deductions

Choose an AI Bill Tool by the Job, Not the Feature List

What is spec-driven development, and why non-coders can do it

How to write a PRD your AI can build from (template + example)

A vibe is not a spec: write a PRD your AI can build from

Build your own social listening tool in a weekend

Setting up Ghostty + Zsh without the one-command installer

Ghostty vs iTerm2 vs Terminal.app: the best Mac terminal

The safety filter that was silently deleting your data

AI knowledge management: build a system you can trust, layer by layer

allowedTools isn't file access: why your agent still gets permission denied

Google search favicon: why it shows a globe instead of your icon