AI's Love Language

Issue #489.May 22, 2026.2 Minute read.
Bytes

Today’s issue: Going to the OpenAI school of naming things, securing Adventure Bay, and agents you can smell from 3 miles away.

Welcome to #489.

Eyeballs logo

The Main Thing

Man holding a boombox over his head

Me telling my agent we're using Zero for this project

AI’s Love Language

This past week, when Chris Tate announced Zero, a love programming language designed for agents, it brought me back to that time in college where I snuck a song I recorded in GarageBand onto a mix CD that I gave to my crush for Valentine’s Day.

I’m not sure if it was the Chad Kroeger style vocals or the 15 layers of acoustic guitar, but we never spoke about it again. Fortunately for Chris, Zero seems a lot more promising than my music career. Here’s a quick overview.

Inspired by Zig and Rust, Zero is designed to make systems programming faster and easier for agents. But what exactly makes a language easier for agents? According to Zero’s design principles, there are several key ingredients:

  • One “obvious” path. Instead of catering to varying human preferences, Zero aims to have a small set of patterns that can be used to accomplish tasks.

  • A robust standard library. The stdlib should cover the vast majority of use cases instead of having to reach for external libraries.

  • Readable tooling. Diagnostics, graphs, facts, size reports, and repair metadata should always be available as structured outputs.

  • Explicit effects. Accessing external systems should always be explicit and easy for the agent to inspect.

Bottom Line: It’s unclear whether an agent-first language like Zero is the future, or if other languages will just build similar features into their runtimes. But regardless, whether it’s for your clanker or your crush, doing things for love is what makes us human 💕.

Upsun logo

Our Friends
(With Benefits)

A man wearing a tie looking shocked

Watching my agent push its "quick refactor" to prod

Your agent needs a sandbox upgrade

It turns out giving your agent unrestricted access to prod is… not a great idea. So before you let Jarvis drop your production database, take a look at Upsun.

It’s a sandbox environment that lets your agent work autonomously on a secure platform that handles container orchestration and isolation.

But what’s the point of security if you can’t be productive? Fortunately, Upsun solves this with a propose-and-test workflow:

  1. Propose: The agent identifies an issue or suggests an optimization.

  2. Clone: Upsun triggers an isolated, byte-level clone of production apps, services, and data in seconds.

  3. Validate: Agents make changes that are easy for humans to validate with live previews.

  4. Review: Once the work is validated, a PR is generated with the changes, ready to be merged into main.

Test it out, for free.

Cool Bits logo

Cool Bits

  1. The Deno team just released Claw Patrol, a security firewall for agents. So whenever there’s a problem around Adventure Bay, Ryan and his team of pups will come to save the day.

  2. Cameron McLoughlin wrote about how to use local models in Zed. It has quirks, but you might save a few shekels.

  3. Blacksmith built a drop-in replacement for GitHub Actions that is 2x faster and 60% cheaper. [sponsored]

  4. Google just open-sourced Ax, a “distributed agent runtime.” If this works anything like the body spray, the agents will smell you coming from 3 miles away.

  5. TanStack Start just added support for deferred hydration, which helps you decrease startup time for large pages.

  6. SurveyJS provides a white-label form builder UI component that integrates with any backend. You can install it from npm, then build dynamic JSON forms directly in your React, Angular, or Vue app with no usage limits. [sponsored]

  7. Expo released version 56 of their SDK, which features performance improvements, React 19.2 support, and Expo UI being ready for production.

  8. Oracle AI Developer Hub is a GitHub repo with technical resources and production-ready patterns for scaling apps using LangChain, LlamaIndex, and high performance vector search. [sponsored]

  9. The npm team added a new feature called staged publishing, which adds an additional approval step before packages go live on the registry.

  10. GitHub Copilot launched its own agentic coding app called “GitHub Copilot app.” Looks like they also went to the OpenAI school of naming things.

  11. The Pierre Computer Co launched DiffsHub, a suspiciously horny-sounding name for a tool that lets you view large GitHub diffs without your browser choking.

  12. After years of unsuccessfully evangelizing to humans, the Chrome team built a modern web guidance skill so your AI can finally build things the right way.