Episode notes
Matt and Kris welcome back Bill Kennedy who's been working on Kronk, AI tooling in Go. They go through what Kronk (and it's dependent library Yzma) is, the power of running models locally, and how they see coding agents and LLMs shaping the future of the industry. We've got supporter content, of course! This week that includes the message-caching trick that turns seconds into thirty milliseconds, why hybrid models quietly wreck the KV cache, Bill's "thirty billion knobs" explainer for dense versus mixture-of-experts models, why demo day is destroying engineering teams, and the…
Chapters
Tap a chapter to play from there.
Fallthrough
by Fallthrough Media · English · Tech & Science
A deep and nuanced conversational podcast focused on technology, software, and computing.
More from Fallthrough
-
E77 · 27 Jun 2026 · 27 min
The Spy Who Banned Me
It's Fallthrough time! In this episode, Kris, Matthew, and Steve get into the time Steve pointed Anthropic's Fable at a side project he hadn't touched since January and watched it shepherd 95 pull requests in three days, the US government export banning Fable and its sibling Mythos, the insanity of Mythos popping a national security agency's boxes in a matter of hours, and why a software engineer that runs 24/7 for anyone with an account is exactly the kind of thing governments get nervous about. We've got supporter content, of course! This week that includes whether your AI agents should…
-
E76 · 20 Jun 2026 · 1 hr 7 min
Of Fables and Mythos
Another week, another Fallthrough! In this episode, Kris and Matthew get into the US government slapping export controls on Anthropic's Mythos and Fable models, the marketing machine that made them sound world-ending, and how much of the AI hype actually holds up. We've got supporter content, of course! This week that includes whether your phone is already an extension of your mind and the "Clean Talk" cleaning influencers who will happily tell you to mix two chemicals that quietly kill you, the Egyptologist on a date who proves a person is smart but people are dumb, why we throw you in jail…
-
E75 · 13 Jun 2026 · 48 min
Shall We Play A Game?
Matt and Kris welcome Fernando Duran, the creator of SAD Servers, the site that drops you onto a broken Linux box over SSH with a timer running and dares you to fix it. They go through what SAD Servers is, the spite-fueled interview story that started it all, how companies use it to hire, and the surprisingly lean infrastructure (plus the tiny Go "checker") that makes the whole thing work. We've got supporter content, of course! This week that includes why technical interviews are broken and the spite that turned one into a business, taking the number two spot on Hacker News without toppling…
-
E73 · 30 May 2026 · 1 hr 1 min
Dependencies All The Way Down
Another week, another Fallthrough! In this episode, Kris and Matthew discuss a personal realization about what AI actually unlocks, their workflow philosophies, local models, a recent hot take from Mitchell Hashimoto, and a thought experiment that would break the NPM ecosystem overnight. Like the previous episode (a pseudo part 1), this episode returns to one question: are you moving in a direction, or just moving fast? Kris argues AI has lifted his coding domain ceiling entirely, while Matthew brings some nuance to the "just submit to the LLM, you're an architect now" take. The duo then…
-
E72 · 23 May 2026 · 33 min
Terminal Velocity
Has AI eaten all of tech? Is 600 billion tokens too many use in a single month? On this week's episode Matt joins Kris to talk about some of the current news in the tech industry. From OpenClaw's maintainer burning through $1.3 million worth of tokens to why we need more epistemic markers, to the rather absurd $725 billion that big tech plans to spend on AI CapEx this year. And of course, there are layoffs coming as well. Support content? We've got a whole bunch of that! This week that includes a ZFS war story where a thousand-fold write amplification bug stalls every file write, the case…
-
E71 · 16 May 2026 · 1 hr 26 min
He Who Controls the Source
This week we've got Kris and Jamie! They open with the Shai-Hulud worm chewing through the npm supply chain and close on the messy economics of who actually pays for open source labor. And there's plenty of great stuff in between: GitHub's everything-platform creep, the case for LLMs as a way out of dependency hell, and the forge alternatives finally maturing into real options. We've got supporter content, of course! This week that includes an expansion on Kris's "pull requests as original sin" theory, the everything-platform rant that compares GitHub to Meta Business Suite, a monologue on…
-
E91 · 3 Oct 2026 · 1 hr 17 min
Keep Your Enemies Closer
Steve is back! He joins Kris and Matt to discuss the drama around DHH's latest keynote at Rails World and Steve's latest views on AI. Interestingly, Steve believes that he'll sit in opposition to how Kris and Matt see the current situation. We start with Steve giving us the long backstory of Rails and DHH. The trio then dives into a conversation about purity tests, whether you can learn from people you dislike, and why making Linux a viable third option for desktop environments is important. From there they dive into whether the agents that carried out hacks were actually told to do that,…
-
E90 · 26 Sep 2026 · 1 hr 8 min
Thinking, Fast and Cheap
Another week, another duo episode, but this time around it's Dylan who's joining Kris! He brings a fresh perspective to the AI conversation that Kris and Matt have been engaging in. The big question: is AI going to kill us all? Dylan thinks the odds are "astronomically small" and neither Kris nor him are going to lose sleep over the fear driven hype. The two are asking bigger questions: whether what language models are doing is thinking, why those who build these models in AI labs aren't actually the people who understand them the best, and whether TypeSafe AI's new Jev model is an LLM and…
-
E89 · 19 Sep 2026 · 1 hr 4 min
Apocalypse Not Now
Kris and Matthew have some more spicy and it's about the latest AI news cycle: a greater than ten percent chance that AI wipes out humanity inside a decade. Neither of them are worried about it, what does is why everyone else is so worried about it. Kris brings his usual devil's advocate perspective: a 10% chance of extinction is also a 90% chance of non-extinction. He then widens the field of view to the more full argument, which is that AI is the easiest thing for everyone to point at. Matt takes a different direction and takes the argument at face value and walks it forward, ending with…
-
E88 · 12 Sep 2026 · 1 hr 1 min
The Elements of Slop
Kris and Matthew have some feelings about AI they want to discuss. No, not about benchmarks or agents, but about the people using it to write things they should be plenty capable of writing themselves: the five paragraph LinkedIn post, the blog post, the tweet. Both Matt and Kris stumbled upon Pangram from the same source: Bryan Cantrill. The two go from there, discussing the ridiculousness of people writing short form posts fully with AI, to the purity tests that people insist on having, to the literacy crisis. This episode has a ton of spice, especially in the expanded edition. Speaking of…
Take it with you
The Melo app keeps playing with the screen off, works in the car and on your watch, wakes you to your station, and browses the whole catalogue offline. Free, no ads, no account.
