AI Native IDE, we make AI-tools and share AI-knowledge
Oslo, Norway
Joined December 2024
- Tweets587
- Following818
- Followers91
- Likes568
Pinned Tweet
prose 0.9.0 just released. It has this new nice dashboard that gives you a summary of your work across sessions.
It also has a feature to do explicit handoff between agent-session (cross agent harness).
npmjs.com/package/@6digit/pr…
11.5h/day is a decent average. What's yours?
Running on 6digit CORDIAL standalone.
We'll be "live-coding" today, Prime Sieve, compiler improvements, etc.
teapaw.org/i/broadcasts/1kKzDDkWM…
6digit studio IDE retweeted
Most of the internet decides who to trust by counting — followers, stars, upvotes, blue checks. But counting is trivially fake; identities are free. We built the whole web on a trust signal that's for sale.
I've been building something that doesn't count. 🧵
I want to do a seminar for AI illiterate people. Let's all build a website or something else that is easy and fun.
If you have any questions, raise your hand so I can tell you to ask the AI.
Musings on agentic development.
All development is gap-closing. If you cannot measure where you are and know where you're going, the LLM cannot effectively reason about whether the gap is being closed or not.
So design your system to be ultra transparent and testable. The more you can float of data to the LLM to ensure correctness, the more it can iterate over. Runnable artifacts are better than prose, often by a lot.
The reason why people are loving Rust for agentic development is that the compiler gives you a literal runnable representation of the gaps that needs to be closed.
If you can model this into your domain so that requirements are also runnable, repeatable and deterministic, you'll be having a better time.
Verification as a runnable artifact = You don't have to verify.
Create these harnesses with agentic coding, have the LLM help you brainstorm and implement it together.
Spend more time on this extended harness than on the core product in the beginning. But always prioritize the extended harness over the shipped artifact; the shipped artifact is literally downstream from your extended harness.
Hope this helps.
Oddly enough, LLMs are pretty bad at prompting. They consistently over-prompt. I think it's in "their nature" to try to "fix the problem" at every step of the way, so prompts are almost always dubiously colored.
Almost more important review generated prompts than code at large.
Google seems to be solving for the wrong problem. If they focused on fast/cheap work-horse models like Gemini 2.5 Flash instead of chasing the high-capability models like Codex/Opus, they would probably see ~15% of my personal agent-interactions, and 99% of my systems traffic.
I guess you can use Gemini 3.5 Flash to make CRUD-services and webpages.
It's actually worse that you still CANNOT *TALK* TO THESE FUCKING MODELS, they just trample ahead like a god damned elephant in a glass factory.
I guess @antigravity is made for CRUD and webpages.
We made an update to CORDIAL:
- Background parallax shader
- Optimistic UI updates
- Snap camera when switching between sessions
- Camera tilt (configurable)
youtu.be/VB6K0A3fsnw
Did we forget to post this here? Reveal of our coding surface CORDIAL.
youtube.com/watch?v=ZVy0vXBH…
Claiming that all languages that compile through LLVM will perform the same is about as dumb as saying all languages that compile to assembly will perform the same.
Do some benchmarking.
6digit prose 0.7.1 is released. Never heard of prose?
Check it out here: npmjs.com/package/@6digit/pr…
6digit studio IDE retweeted
A TUI is an evolutionary dead-end and we'll look back at this time in history as a mass hallucination where everybody thought they were Neo because they could operate a terminal.
Mnemonics-package just released:
npmjs.com/package/@6digit/mn…
Resolves symbols recursively under a reader monad.