Pinned Tweet
Excited to unveil Bonsai today. It’s been a blast to build and I’m really proud of how it’s turned out. Internally it’s the only tool we use for work. We hope you enjoy the experience!
Spot on
We only do a tiny fraction of the interesting, valuable things there are to do in the world.
We're limited by time, by energy, by how much context we can keep in our brains. We do what we can, but the world has much more capacity for beauty and for craft.
To me, this is what the manifest destiny of technology is about: unleashing humanity's potential by increasing our leverage to do our work. Our hands must reach further and do more.
The current AI era has given us cognitive leverage. We can now pay perfect attention to every problem we can explain to an LLM.
Last week, every engineer on our team shipped a billion tokens per day. That's 8,000 novels' worth of tokens per person, every day. We aren't slop cannons. We aren't lost to cyberpsychosis (although I now dream in neuralese). We got here by pushing the limits of cognition inside Yak Labs.
Meanwhile, I have old coworkers who call autocomplete in their IDE slop. The average knowledge worker is using LLMs as a fancy Google. They can't get a consistently good PowerPoint out of an LLM.
The gap between pioneers and these users is growing.
It worries me. We have in our hands the most powerful contribution to human flourishing we've ever created. But all of that leverage is being left on the table. It requires you to become a pioneer to use it.
No one should have to learn how to become a pioneer. Instead of mastering agent swarms, they should be mastering their craft and finding their own opportunities for leverage.
That's why we're building Bonsai.
Bonsai lives on your computer and learns every corner of your role, your team and your company. It's like having a little me at your desk, figuring out every night how to change your AI setup so you have more leverage tomorrow.
You don't need your entire team on Bonsai for it to work. Our goal is to make YOU better, TODAY. We've learned from running AI enablement for hundreds of employees: top-down mandates don't work. "Use more tokens" !== "Use tokens well". Your top performers will zoom ahead to the frontier and leave everyone else behind.
Instead, Bonsai starts with one person. The frontier will spread one desk at a time. Bonsai will help you keep the work you love and do more of it. Hand the swarm everything below your bar. Go find the rest of those interesting, valuable things to do. Cultivate your craft; Bonsai will cultivate your swarm.
My DMs are open. I'd love to hear from you.
Alex Goddijn retweeted
Excited to finally get to share what we’ve been working on! This is only the beginning, we hope you enjoy.
Alex Goddijn retweeted
Having started a company <2 months ago I've been asking myself how much faster it actually is building with LLMs.
Using stats from Supabase, Tailscale, and PostHog in '21 as our comp, we're (conservatively) around 30x more productive per engineer.
What a time to be alive.
Finally at a point where Fable it admits it also has not read the 3k loc of tests, but is pretty confident they're good to ship 🚢
When someone asks me what the first signs were that AI was becoming sentient I'll point to this tweet
Dev flow getting pretty dialled now. Working on the dev loops and seeing them compound is pretty addicting...
Agents tend to go for the quick win over the well architected feature (which makes sense given how they post train).
They're closer to hackers than software engineers. We've been experimenting with cheap ways to move the balance closer to "think" before "act".
Here's a skill we developed through trial and error that I find myself using often these days (link in comments).
BLUF: Before it writes code, it drafts the architecture, tries to break it, challenges the assumptions, and reviews it from a few angles, then gives a keep / revise / redesign call. That helps surface the wrong design up front instead of one bug at a time.
I've been teaching since I was young.
I started with youtube tutorials on how to run private world of warcraft servers at 11, teaching Java at 14, then maths, physics, chemistry, and comp sci from high school through to university.
At Palantir, my role as a forward deployed engineer was also essentially to teach people how to configure and use the platform, turning it into the software equivalent of an iron man suit.
It's always been gratifying to empower people through teaching, but I mostly did it out of frustration that things are actually a lot simpler than most people make them out to be.
And today it's the same for using AI.
AI pilled devs are creating a whole new jargon around AI tooling, an leveraging the technology a lot better than non technical folks, leaving them feeling out of the loop.
So my next quest is to teach AI to teach people how to use AI better. I'll keep you posted on how that's going.












