Jonathan Levine

Eng Phys @Carleton_U

[Jonathan Levine @TheJLevine]

@Chris_Wozniczek I reject the premise. that's scarcity mindset. I want both.

[Jonathan Levine @TheJLevine]

@yminsky Agreed and I have been having A LOT of fun vibe coding with bonsai_term my own terminal app to recover tmux sessions. Really love the tooling.
https://t.co/qkLtB1rcD4

[Jonathan Levine @TheJLevine]

The question is. What will you build? Anon.

[Seve @seveibar]

a lot of people (my angels) are asking: what does it mean that OpenAI can now layout and route PCBs with astra? am i cooked? (1/6)

[Jonathan Levine @TheJLevine]

@alexocheema Yea! Owen 3.8 27b Q4 has been great. Basically wrote this tool https://t.co/qkLtB1rcD4 entirely with Owen. works great for me.

[Jonathan Levine @TheJLevine]

@jaga_prasanna Fascinating we see this trend on GPUs in the co text of LLM serving. I noticed this when running simulations on CPU a few years ago!
https://t.co/yaDSdb0YSR

[Jonathan Levine @TheJLevine]

trying out a new flow where I have codex making the plan / architect the solution then handoff to qwen 3.8 for the implementation. seems to be working well. What are you guys doing?

[Jonathan Levine @TheJLevine]

@jon3k Long time @AlmaLinux user (yes on the desktop and not on the server lol) and switched to @OmarchyLinux a year ago. Loved the aesthetic and window management. Now alma is purely on the server. Credit where it’s due though, @AlmaLinux has been my most stable machine 🚀

[Jonathan Levine @TheJLevine]

@andres_vidoza Install @OmarchyLinux! I promise you won't look back.

[Jonathan Levine @TheJLevine]

@NojanSheybani Personally I think the big EDAs are quite well positioned for the agentic era. Trends we are seeing is that agents love to operate at the cli level and the big EDAs have amazing clis. Electrical engineering is going to go parabolic 🚀🚀🚀

[Jonathan Levine @TheJLevine]

@AryaTschand https://t.co/jgxlTG7AnN

[Jonathan Levine @TheJLevine]

@joefioti The theme I’ve noticed from hot chips is that almost the LLMs are becoming the compilers and the hardware designers. Do you think the age of the classical compiler is coming to an end? Or stacks like @luminal_ai and @__tinygrad__ will be the platforms on which the LLM optimizes?

[Jonathan Levine @TheJLevine]

“There are decades where nothing happens and weeks where decades happen.” @dhh with the banger. This will be the theme of the second half of the 2020s. Accelerate. We can build anything.

[Lex Fridman @lexfridman]

Here's my conversation with @DHH. He is back for round 2! It was an epic fun 5 hour conversation about the future of programming, AI, Linux, and human civilization.

It's here on X in full and is up everywhere else (see comment).

Timestamps:
0:00 - Episode highlight
1:27 - Introduction
2:56 - Programming with AI agents
18:14 - How software will change
27:30 - AI impact on open source
37:21 - Building Omarchy Linux distro
47:05 - Vibe coding vs agentic engineering
1:00:06 - The end of manual programming
1:10:24 - Advice for programmers
1:22:30 - Surviving Internet Hate
1:31:46 - Programming setup for AI Agents
1:44:11 - Obsessing about speed
2:07:06 - Voice prompting vs typing
2:21:05 - Best AI coding models
2:37:55 - Best AI coding harnesses
2:50:57 - AI video generation and filmmaking
3:10:28 - Fatherhood
3:38:35 - Linux will win the desktop
3:49:51 - PewDiePie
3:59:24 - Future of programming
4:22:17 - Politics and immigration
4:53:54 - Longevity, over-optimization, and fear of death
5:05:38 - Eternal recurrence and future of human civization

[Jonathan Levine @TheJLevine]

@bubbleboi Is that not google’s HLS stack? https://t.co/a5FoBjvmok

[Jonathan Levine @TheJLevine]

@paulg I have found that basically my parents were right about everything. Maybe I was lucky?

[Jonathan Levine @TheJLevine]

@EtherCoins @danieltvela hmmm. I am running Pi as the harness. I just hit 40 toks/sec on a fresh chat. Are you hitting those speeds consistently even as context fills in the chat?

[Jonathan Levine @TheJLevine]

@EtherCoins @danieltvela Wait is my speed slow? I get BIG drop in perf though as ctx builds. Down to 15 toks/sec now! running a 12900k with 96GB DDR5 @ 4800 MT/s. qwen3.8 27b @ 4 bit quant.

[Jonathan Levine @TheJLevine]

@ico_TC I picked up a second-hand Altera PCIe card, and within a couple of hours I had my own custom accelerator running on it with bidirectional PCIe DMA. Codex wrote the FPGA integration, Linux driver, and C++/Python host interfaces. Hardware design is going parabolic.

[Jonathan Levine @TheJLevine]

imagine being a windows user now. imagine. Thank you @dhh for the beautiful os! reviving my 10 year old machine!

[Jonathan Levine @TheJLevine]

Ok qwen 3.8 27b is really good. Fixed a bug in tmux-recovery that none of the other models could. This reveals an inportant distinction in the age of agentic engineering (ascension). It’s important to have your own internal benchmarks. What metrics are you using?