1046: The Pi Updates are insane

Summary of 1046: The Pi Updates are insane

by Wes Bos & Scott Tolinski - Full Stack JavaScript Web Developers

1h 40m•October 12, 2026
Listen to Episode

Overview of 1046: The Pi Updates are insane

This episode is a rapid-fire rundown of a wild week in developer tooling and AI: the hosts cover everything from a facelift for example.com, the growing “death of pull requests” because of AI-generated slop, and major releases like Pi 1.0 / Pi Durable, SvelteKit, Effect 4, Screen Studio 4, and a wave of AI announcements from OpenAI, Cloudflare, Google, Claude, and more. The throughline is clear: software is getting more agentic, more customizable, and more opinionated about quality as low-effort AI code floods the ecosystem.

Major Web Dev and OSS Updates

example.com got an update

  • The long-standing placeholder domain example.com received a facelift.
  • It’s a real domain owned by IANA and widely used in docs, tests, and examples.
  • The hosts joked about how many CI systems and health checks may depend on it.

The “death of pull requests”

  • Several major maintainers are turning off external PRs because of AI-generated contribution spam.
  • The main complaint isn’t just code quality—it’s the flood of low-context, low-effort PRs.
  • The shift is toward:
    • issue-first discussions
    • clearer problem statements
    • maintainers using their own agents to implement changes
  • Takeaway: AI may be making code cheaper to generate, but it’s also making trust and review more expensive.

Sindre Sorhus and project overload

  • The episode highlights maintainers like Sindre Sorhus and others who are overwhelmed by AI-generated PRs.
  • The hosts emphasize how hard it is to sustain open source projects when everyone just forks instead of collaborating upstream.

Pi 1.0 and Pi Durable

Pi 1.0 is a major release

  • Pi is described as a minimal, extensible agent harness.
  • New additions include:
    • MCP support built in
    • code mode
    • deferred tool loading
    • caching improvements
    • better theme/fullscreen behavior

Why “code mode” matters

  • Instead of the agent calling tools one by one and bloating context, Pi can:
    • generate a small program
    • run it in a sandbox
    • return only the useful results
  • This is framed as a big win for:
    • context efficiency
    • tool parallelism
    • faster agent execution

Pi Durable

  • A new long-running agent harness for resumable, replay-safe workflows.
  • Supports:
    • stored conversation history
    • task recovery after crashes
    • forking and parallel work
    • multiplayer/shared agent sessions
  • The hosts see it as a strong base for:
    • internal assistant systems
    • customer-facing agent products
    • long-lived automations that can survive failures

Terseo Joins Supabase

A database platform merger

  • Turso is joining Supabase.
  • The practical result: one place for hosted:
    • Postgres
    • SQLite / LibSQL
  • The hosts view this as a net positive for developers who want a simpler “one-stop shop” for database hosting.
  • They speculate this could eventually lead to more SQLite-like options inside the Supabase ecosystem.

Screen Studio 4 and the rise of high-quality software

Screen Studio 4 release

  • The app, known for polished screen recordings and demo videos, shipped a big update.
  • New features include:
    • glass-like visual effects
    • camera cutout editing
    • transcript-based editing
    • face tracking
    • split views
    • sharper zooming / text rendering

Why it stood out

  • The hosts praised the team for not just cloning the cheap AI-generated alternatives.
  • Instead, they doubled down on quality and solved hard problems like:
    • making zoomed text stay crisp
    • improving overall polish and usability
  • Main takeaway: in a world full of “slop forks,” strong product taste still wins.

SvelteKit, Effect, and Alchemy

SvelteKit updates

  • The episode covers a large SvelteKit release with improvements like:
    • less config-file overhead
    • better env var validation
    • simpler service workers
    • better instrumentation support
    • remote functions progressing, though not fully ready yet

Effect 4

  • Effect is getting more popular and the hosts note it has a steeper mental model.
  • Version 4 brings:
    • smaller bundles
    • much better concurrency
    • lower memory usage
    • a dependency-free core
    • LTS support
  • They discuss how Effect feels intimidating at first, but increasingly worth it.

Alchemy

  • Alchemy is introduced as an infrastructure-as-code layer the hosts are liking.
  • It helps define and connect cloud resources in a more deterministic way.
  • Useful for people wanting portability across providers like:
    • Cloudflare
    • AWS
    • Fly
    • Railway
    • Neon
    • PlanetScale
  • The hosts like the idea of avoiding lock-in while still enjoying TypeScript-based infra config.

Cloudflare config

  • Cloudflare also introduced a new cloudflare.config.ts approach.
  • The point is to make infrastructure config more programmable than static JSON/TOML-style files.

OpenAI Dev Day and the AI Platform Wars

OpenAI “Dots”

  • OpenAI introduced Dots, effectively always-on agent computers.
  • They run in a VM and can install/run apps.
  • Seen as a response to similar agent device efforts from Meta.

GPT-6.1 “Soul” / Luna-related discussion

  • The hosts discuss OpenAI’s newer model releases and their mixed feelings.
  • Main criticism:
    • OpenAI models still feel over-engineered and less pleasant to work with than some competitors.
    • Great for quick prototypes, less pleasant for developer-heavy workflows.

OpenAI extensions

  • OpenAI is pushing richer app integrations inside ChatGPT:
    • persistent sidebars
    • app-like experiences
    • UI that updates over time instead of regenerating every turn
  • They also mentioned:
    • Spaces for team collaboration
    • Pages
    • Slides
    • Tasks
    • Sites for vibe-coded mini web apps inside ChatGPT

Cloudflare’s “Clef”

  • Cloudflare released Clef, a decision model similar to JEV-style classification models.
  • It’s positioned as an alternative for fast, structured, low-cost decisions.
  • The hosts note:
    • it appears somewhat stronger on some benchmarks
    • but is also more expensive
    • and has a larger context window

Other AI model and platform news

Google Gemini 4 “Argon”

  • A new Gemini frontier model is in limited testing.
  • Early talk suggests it may be competitive with or ahead of some leading models.
  • The hosts think Google has had time to cook and may be ready to re-enter the race seriously.

Black Forest Labs Flux 3

  • Flux 3 adds powerful image editing features:
    • regenerate specific regions of an image
    • use reference images
    • improve single-prompt quality
  • The hosts see this as a serious competitor to Midjourney and other image tools.

Claude mods

  • Claude now supports mods written in TypeScript.
  • These can customize:
    • the UI
    • agent behavior
    • extra panels / status widgets
  • Example mods include weather panels and replay history visualizations.

Muse gadgets

  • Meta’s Muse agent gets hardware extensions via ESP32/Raspberry Pi-style devices.
  • The hosts are excited about hacking physical devices into agent workflows.
  • This aligns with their own work on agent-connected hardware and “prompt boy”-style systems.

NVIDIA Lyra 2.0

  • A world model that can generate a 3D walkable scene from a single 2D image.
  • The hosts are excited about the implications for:
    • virtual reality
    • real estate tours
    • game-like environments
    • spatial computing
  • They connect this to future home scans, VR walkthroughs, and richer digital twins.

JavaScript and language-level news

Composite equality proposal

  • A Stage 2 TC39 proposal would allow composite equality for objects/arrays/maps/sets.
  • The goal is to make value-based comparison more natural instead of relying on:
    • JSON stringification hacks
    • third-party deep equality helpers
  • It’s early, but the hosts are optimistic.

Culture and hardware side quests

Pigeon RFC auction

  • They mention RFC 1149, the famous joke standard for transmitting IP datagrams over pigeons.
  • A real-world pigeon-delivered packet from 2001 is up for auction at Christie’s.
  • The hosts find it funny that a joke internet artifact has become a collectible.

Recompiling and decompiling games

  • The episode highlights the growing community around recompiling classic games.
  • Benefits include:
    • better performance
    • native ports
    • modding
    • fixes for old bugs
  • They mention games being modernized, remixed, or ported across platforms via recompilation.

Practical takeaways

For OSS maintainers

  • Consider issue-first workflows to reduce PR spam.
  • Use agents internally to implement changes after discussion.
  • Be selective about what outside contributions you accept.

For AI app builders

  • Favor durable, resumable agent systems for long-running workflows.
  • Keep context slim and push work into code/sandboxed execution where possible.
  • Expect more platform bundling from the big AI vendors.

For frontend and tooling teams

  • The episode’s broader message: high quality still matters.
  • Teams that ship polish, good defaults, and thoughtful UX can stand out even in an age of easy cloning.

Closing note

  • The episode ends by plugging the Syntax meetup in San Francisco on October 27th, 5–7 PM, and encouraging listeners to grab tickets while they’re still available.