Whatβs Prompt API?
Whatβs Prompt API?
It has been so much fun making Claude Code edit videos via Remotion. React is all you need to know to be a video creator. Hereβs a short I made today
youtube.com/shorts/EelbO...
Ah. Then MBPro it is. The hardware is great, just get as much memory as possible. I couldnβt afford more than 16GB, but my friend got ~96GB and can run almost any model
I will always recommend a Mac Mini. Been using one way before the claw wave started. It is usually 3x cheaper than the equivalent MacBook. Mine has the M4 chip, I have run many models locally.
Not only is it normal, it is usually a sign that you are learning something. This is what I have seen for myself in my two decades of software engineering experience.
Of course, if this feeling is so strong that it makes you uncomfortable, then something needs to be done.
It used to be %SPECIFIC_EXAMPLE% before LLMs arrived
One more idea - how about using CSS in the 3-d space. I have no idea whether it works with world models
For example - why not CSS for books? Two months ago I was trying to create books using CSS - I am not good at writing or at CSS so that effort likely will take time.
Great point.
Also, I think that the craftsmanship never dies, it evolves. For example, there might be way more new use cases of CSS and HTML than we can imagine now.
I asked it how it did it, and thatβs how I know. I promptly asked it to turn it into a skill, which it did.
None of this is new - @steipete.me had the exact same experience a few months ago. And I knew it. But it is still a nice experience to see that these systems can learn
Tried to talk to my tinyclaw bot by sending a voice message. It asked me for the location to whisper. When I said go figure out something local, it created a python environment, installed whisper and transcribed the audio
I plan to collect them and write a blog post.
The coding agent I built (vibe coded) is github.com/debamitro/on... .
I have been trying to use it on existing codebases, using Claude as the LLM. It is definitely a year behind Claude Code
The more I try to use the coding agent I built, the more I realize the importance of the system prompt. I am using Claude Sonnet and yet it is 10x dumber than the likes of Claude Code.
My blog post describing how I started playing with TinyClaw over Telegram.
debamitro.github.io/blog/riding-...
Listened to @steipete.meβs interview with @lexfridman.bsky.social. Itβs more important than ever to use AI agents. Itβs equally important to be intensely human ourselves. With all our imperfections. And have fun with whatever weβre doing.
Created my first telegram bot and connected it to a tinyclaw running on my Mac Mini. This interface feels so new to me!
At the same time it is quite revolutionary. Why didnβt I know about telegram bots before?
Just paste this post into Claude Code, from a directory with the assets. Should work
Hackathons of the future will become spec-a-thons.
Only the spec matters now - code is a commodity.
Save your specs, lock them up!
Interesting β¦
Is making apps for AI bots going to be the next big thing?
I heard this name after a very long time! I am surprised that it has survived. Everyone I know uses Claude Code or any of their identical competitors
Maybe because a lot usage in USA is by startups and hackathon participants? Clerk is super useful for even the tiniest web app. WorkOS looks like what enterprises need
Whatβs an example?
Great article, three years old, but still relevant as the manual for making autonomous agents. Written by the guy who built moltbook www.mattprd.com/p/the-comple...
Last night I finally had that βa haβ moment in my mind - when I finally realized how agent skills work. And how the basic concept can work with any agent.
Even the agents built before Claude Code arrived on the scene
I never knew I could ask a command-line tool for a travel plan. It searched the web and came out with some recommendations.
It's so cool that you can do such things
Hereβs my recommendation: use an agent to build a small agent and see how it works. I just did this yesterday, use and throw stuff. Documented at debamitro.github.io/blog/creatin...
Today I used a coding agent to create a new coding agent. Feeling very excited to see this happen in front of my eyes! I didnβt write a single line of code yet.
Naive question here: how is the authentication handled - API keys or passwords
This is the best! I have seen so many projects die because leadership didnβt allow them to prosper. Iβll add this to my list of predictions about software in the age of AI