Turning vibe coders into software engineers, one episode at a time. We share our learnings on how AI can augmenting developer, not
replacing their craft.
This week @kaushikgopal and I had the pleasure to chat @mitchellh on the pod๐๏ธ!
Refreshing to hear someone of his caliber bring such a grounded perspective to agentic coding.
We also talked about Ghostty, and how terminal performance gains make tools like Claude Code possible. (He even explains what's behind scrollback perf issues).
A lot of gems in this one. Check it out!
The calendar alerts he sets for warm starts. Hail Mary prompts across GitHub issues. Why AI leaderboards are BS and rarely translate to value. A vibecoder he respects who's actually shipped.
Listen to that and more ๐
https://t.co/bi7OJfdCi8
๐๏ธ Our first guest in the AI series is the legend @mitchellh ๐๏ธ
We covered a lot of ground and learned a tonne from him: ๐ป Ghostty's internals and why tmux & certain shells strip its best features. His path from AI skeptic to cautious optimist.
We're going to be releasing ep 310 (our 10th episode in the new series) in the next day or two.
๐๏ธ it's an ultra special one cause we're starting up the guest series again... and boy do we have a guest for you.
stay tuned!
Karpathy's goal for agents: maximize how long they run without intervention.
While individual speed is going up, "system" speed isn't (the false summit โฐ๏ธ).
In ep 309 we talk about resolving this tension with Background Agents!
Our LLM explainer (ep 303) got pretty popular, so we thought we'll follow up with the natural progression - how do image models work - in 20 minutes!
also... what does Michalangelo have to do, with understanding stable diffusion models? ๐ค
@kaushikgopal & @IurySza dug into:
- shaping the agent environment & 5 themes for harness engineering
- then on building a harness
For your ears ๐ง
https://t.co/G5y0ExEp1x
OpenAI spent 20% of every week cleaning up AI slop from their own coding agents. So they built a "harness" shaping the agent's environment so the slop never ships in the first place.
New episode 307 on Harness Engineering !
Are AGENTS[.]md files even useful? Thereโs finally a paper asking the obvious question.
@kaushikgopal and @IurySza get into the paper, the sync problem, and the small amount of discipline required to stop your agent setup from turning into a mess.
No episode this week dear listeners but fun newsletter with that story and 10 promised tips from @IurySza and @kaushikgopal
https://t.co/e8yjtQBNI7
Subscribe to our newsletter for more of this content!
We almost shipped an episode with a silent master track ๐ฌ
Adobe Podcast returned 58 minutes of completely silent video for one of our episodes. No audio. Just two guys miming a podcast. ๐จ
Here's how we recovered it ๐งต
A script that analyzed mouth movements frame-by-frame from the silent video, correlated that motion signal against audio energy bursts in our local recordings, and recovered the exact timing offsets.
... only felt like this simplest solution ๐
If you're curious to learn more about subagents, new @FragmentedCast ep out with @kaushikgopal & @IurySza!
what subagents actually are, a convenient RTS scout analogy, when to spawn them vs when not to. all in this week's episode!
https://t.co/UCevKBz1No
Master subagents and your AI coding sessions stay sharp for hours with clean results and no context pollution.
Under the hood though? they're just a tool call. One model prompting another in a scoped context. ๐ก
thanks for sharing our episode last weekend @androidweekly ๐
Ep 303 explains how LLMs work in a simple and relatable way. It really is one we'd love all devs to listen to.
Skills solve context bloat โ every instruction in AGENTS eats tokens until your agent's results tank.
We also cover the dark side of skills.
When to use skills, how they differ from commands/MCPs, and how not to get pwned:
๐ง https://t.co/s7rAs9ODMa
Everyone's talking about @openclaw (Clawdbot) right now.
But have you looked at what actually makes it so extensible? Agent Skills!
modular SKILL md files that teach an agent how to do anything.
We did a deep dive on Ep 304. ๐งต