Get live statistics and analysis of Will Bunting's profile on X / Twitter

Harsh reality when building systems: Most problems come from interactions between parts. Every new piece you add increases complexity quadratically. Simplify ruthlessly. Resist external pressure to add parts.

There is no such thing as “value-free AI.” Every system reflects culture, history, and the environment of its creators. The danger isn’t morality vs. no morality. The danger is pretending neutrality exists.

With my coding environment it's always a feeling of: *"If I just fix this one thing THEN it will be perfect and I'm done"* Famous last words...

I had the experience of feeling sorry for an agent today... I stopped it after it succeeded but before it ran the validations So it never knew after all its hard work that it actually did get the right answer

Pre-AI coding mistake: Picking off easy peripheral tasks first. Post-AI coding mistake: Leaving a single blocker in place while your agents sit idle. Clear bottlenecks first. Always.

My dev setup used to be calcifying — small tweaks, slow changes. Now it’s fluid. AI-driven workflows shift fast, and each change can bring a big productivity jump if you adapt quickly.

I’ve always tracked my work. But kicking off AI agents from Linear forces me to do it every time — and with more detail. The result: a consistent record of next steps, decisions, and context. Better for me. Clearer for the team.

Wild thought: Social media didn't make leaders kinder, it just made aggression expensive. Apple can't do dramatic firings. Sports coaches can’t berate rookies. The Yakuza can't recruit Gen Z.

In the early 2000s, releasing software had a quality floor: You needed physical discs, manufacturing, and distribution with no "over the air" fixes. Now, digital releases have pushed the quality floor to near zero. No wonder we are flooded with buggy software.

I spent years chasing 1% productivity gains in my dev setup. One line of neovim config here -- a bitter keyboard there. Now I’m getting 10% gains overnight with every new AI workflow adjustment. Totally different ballgame

The hardest part of any coding task isn’t finishing it. It’s starting it. AI agents erase the blank page problem. Dictate a task, get a PR back, and start editing. Activation energy: gone.

Photographic memory was never considered rude. Perfect audio recall via AI pendant somehow is. The inconsistency reveals we're not ready for augmented cognition.

Successful people excel at getting plates spinning in the air (0 to 1) and keeping them there (systems). • Initial burst of effort is necessary to get things moving • Once in motion, use systems to maintain progress Push hard, get projects going, then let systems keep them up

We could save an entire generation from brain rot... Turn schools into giant Faraday cages. No WiFi. No 5G. You want the internet? Go to the computer lab. Connect via ethernet.

Even if it is not the most "productive"... You'll learn a ton by trying to write a fullstack app in C. It teaches you how the computer actually works instead of teaching you how a bunch of abstraction layers work.

After having kids, one realizes that more can be done in less time than one thinks. Time constraints = productivity. During COVID, with endless hours, I squandered work time. Now, with a two year old, I'm hyper-aware of every minute.

In the early 2000s, releasing software had a quality floor: You needed physical discs, manufacturing, and distribution with no "over the air" fixes. Now, digital releases have pushed the quality floor to near zero. No wonder we are flooded with buggy software.

Am I the only one who still misses RTS games? When Warcraft IV was never announced, something in the industry shifted. The golden age of strategy games felt like it ended, and it’s a void that’s never been filled.

Current AI feels fundamentally different from human thought. We evolved complex reasoning and spoken languages skills without any written text for tens of thousands of years. There's a "mentalese" that predates written language — and LLMs don't have it.

The hardest part of any coding task isn’t finishing it. It’s starting it. AI agents erase the blank page problem. Dictate a task, get a PR back, and start editing. Activation energy: gone.

Harsh reality when building systems: Most problems come from interactions between parts. Every new piece you add increases complexity quadratically. Simplify ruthlessly. Resist external pressure to add parts.

Problem solving does not mean perfection. A healthy obsession over quality in what you ship is critical for customers being truly satisfied But obsess too much and you will have no customers (never shipped) Balance is key.

What is the best habit you’ve built this year? For me, it has been: Consistent Japanese immersion: whether passive or active Why? - I’m always progressing - I can do it while handsfree (with podcasts, or TV audio) - Exposure to different worlds So simple but so powerful.

These three principles from Steve Jobs' product development process can transform your approach to creating outstanding products These insights are more enlightening than any book I've ever read on the subject Here are his words 👇

Moore v. United States could prevent startups from paying younger employees with equity. Imagine having shares you can't sell, an IPO years away, and no cash to pay taxes on unrealized gains.

The First-Time Japanese-Learner Playbook. (Read this if you want to learn fast, enjoy the process, and persist to the finish line.)

Am I alone in preferring long files for code rather than splitting to many files? My Neovim configuration is one giant init.lua

A commonly given financial advice is to save x% of your salary and rely on compounding for growth But for many young people their salaries are too low compared to cost of living for compounding to make a difference That money should be invested in their skill growth instead

I've switched note taking systems a bunch of times and still struggle to not go crazy I just want: 1. vim motions 2. backlinking 3. Timeblocking / agenda / todo tracking Org-roam is the best thing I've found so far, but Emacs drives me nuts Anyone suggestions?

My dev setup used to be calcifying — small tweaks, slow changes. Now it’s fluid. AI-driven workflows shift fast, and each change can bring a big productivity jump if you adapt quickly.

I had the experience of feeling sorry for an agent today... I stopped it after it succeeded but before it ran the validations So it never knew after all its hard work that it actually did get the right answer

Dream Big What if we had a Hyperloop from Tokyo to California across the Bering Strait? 6 hours to SF. No crazy airport security. I'm sure getting Russia and North Korea on board will be no trouble...