The Next Token

Dillon Mulroy, Rhys Sullivan, and Sunil Pai

The Next Token is a technical, conversational podcast from Sunil Pai, Dillon Mulroy, and Rhys Sullivan about the ideas, tools, and weird edges shaping software right now. Expect candid conversations about AI, developer experience, building on the web, open source, infrastructure, and the work behind the work: what they are trying, what is breaking, what is actually useful, and what deserves a second look. Cloudflare sponsors the show, but the conversations are host-led, curious, and independent in spirit.

Episodios

  1. Ep 0: "None of My Software Has Gotten Better"

    hace 4 días

    Ep 0: "None of My Software Has Gotten Better"

    Recorded June 15, 2026. Dillon, Rhys, and Sunil discuss the Fable/Mythos situation and require receipts for bold claims about how AI is changing everything. Follow Dillon Mulroy - @dillon_mulroy / dillonis.online / twitch.tv/dillon Rhys Sullivan - @RhysSullivan / rhys.dev / executor.sh Sunil Pai - @threepointone / sunilpai.dev The Next Token - YouTube / X Topics The Fable/Mythos situation Feelings, etc. Can you trust a tool that quietly degrades itself? Why more code hasn't meant better software. Open weights, sovereign models, and where access to intelligence is heading. Receipts Required Revisiting bold predictions (accurate and otherwise). In the News Claude Fable 5 & Claude Mythos 5. Anthropic's launch announcement (June 9, 2026): anthropic.com/news/claude-fable-5-mythos-5. Priced at $10 per million input tokens / $50 per million output tokens. Safeguards fall back to Claude Opus 4.8 on: cybersecurity, biology, chemistry, and distillation. Access suspended by the US government. Anthropic's statement (June 12, 2026): anthropic.com/news/fable-mythos-access. Fable 5 / Mythos 5 system card (safeguards, alignment, evals): anthropic.com/claude-fable-5-mythos-5-system-card. Project Glasswing (the Mythos cybersecurity project): anthropic.com/glasswing. Receipts Required Geoffrey Hinton: "People should stop training radiologists now. It's just completely obvious that within five years deep learning is going to do better than radiologists. … It might be ten years." (Machine Learning and the Market for Intelligence, 2016) Dario Amodei (Anthropic): "I think we will be there in three to six months, where AI is writing 90% of the code. And then, in 12 months, we may be in a world where AI is writing essentially all of the code." (Council on Foreign Relations, March 10, 2025) Sundar Pichai (Google): "Today, more than a quarter of all new code at Google is generated by AI, then reviewed and accepted by engineers. This helps our engineers do more and move faster." (Q3 2024 earnings call, October 2024) "We've been using AI to generate code internally at Google for a while. Today, 75% of all new code at Google is now AI-generated and approved by engineers, up from 50% last fall." (Google Cloud blog post, April 2026) Boris Cherny (Claude Code, Anthropic): "I don't prompt Claude anymore. I have loops that are running. They're the ones that are prompting Claude and figuring out what to do. My job is to write loops." (Source, June 2026) Peter Steinberger (OpenClaw, OpenAI): "Here's your monthly reminder that you shouldn't be prompting coding agents anymore. You should be designing loops that prompt your agents." (X, June 7, 2026.) Tweets & Code Referenced Dillon's Fable tweet "imagine if they banned all models and we all went back to writing code by hand and the last 6 months were just a fever dream…imagine" (X, June 12, 2026) Greg Brockman's tweet about the old days "how did we ever write all that code by hand" (X, February 13, 2026) Sunil's "$250 run" tweet "spent all day on fable for a giant PR. ~10kloc, lots of testing and intervention. 250$. I... don't think it's worth it? happy with 4.8/5.5, and the quality of work is better when it's smaler steps. Still rocking @cursor_ai, that's software that I still love using on the daily." (X, June 11, 2026) Mitchell Hashimoto's perf improvements tweet "I've got an agent in a loop optimizing a renderer with the goal to minimize frame times (and tests to measure). It got times down from 88ms to 2ms and allocations down from ~150K to 500. Sounds good, right? Wrong. This is exactly why agent psychosis is a big fucking problem. …" (X, May 28, 2026) shadcn's /improve skill "​​Survey any codebase as a senior advisor and produce prioritized, self-contained implementation plans for OTHER models/agents to execute. Strictly read-only on source code — never implements, fixes, or refactors anything itself. Use when asked to audit a codebase, find improvement opportunities (bugs, security, performance, test coverage, tech debt, migrations, DX), suggest features or where to take the project next (roadmap, product direction), or generate handoff plans for another agent to implement." Factorio "Factorio is a game in which you build and maintain factories." Books Referenced Where Good Ideas Come From by Steven Johnson (also: Sunil's blog post about this) PGP Source Code and Internals by Philip Zimmermann In Person AI Engineer World's Fair. June 29-July 2, 2026, San Francisco. Coffeeflare. Free coffee meetup with the team during the Fair (no badge needed) - Tuesday, June 30, 2026, 7am-10am, Sightglass Coffee, San Francisco. Credits The Next Token is sponsored by Cloudflare. Subscribe on YouTube: @NextTokenShow and wherever you get your podcasts.

    52 min

Acerca de

The Next Token is a technical, conversational podcast from Sunil Pai, Dillon Mulroy, and Rhys Sullivan about the ideas, tools, and weird edges shaping software right now. Expect candid conversations about AI, developer experience, building on the web, open source, infrastructure, and the work behind the work: what they are trying, what is breaking, what is actually useful, and what deserves a second look. Cloudflare sponsors the show, but the conversations are host-led, curious, and independent in spirit.