Issue #2 — Agents earn their keep

Sun, 05 Jul 2026 00:00:00 GMT

Week two, and the labs came out swinging — Anthropic shipped twice in one day and Google put computer use in its fast model. Meanwhile the trending pages tell their own story: security agents and agent multiplexers. The agents aren't the demo anymore; they're the workforce.

The tools

★ strix (agents, OSS)
Open-source AI pentesting agent — 37K stars — that probes your app for vulnerabilities before someone else does.
— Attackers automated first; this is the cheapest way to make sure your side is automated too.

herdr (agents, OSS)
Agent multiplexer for your terminal — run and juggle several AI agents in parallel (12K stars).
— If you run more than one agent, tab chaos is real — this is tmux energy for the agent era.

design.md (coding, OSS)
Google's format spec for describing visual identity and design rules to AI coding agents.
— Your agent stops guessing your brand colours — it's a README for taste, and it's about time.

orca (agents, OSS)
Desktop environment for running multiple AI coding agents in parallel, on your own subscriptions.
— herdr with a GUI — when the same idea trends twice in one week, the pattern is the story.

Worth reading

What Hacker News Gets Right About AI Coding Agents in 2026
— The argument moved from 'is this real' to 'how do we make it economical and repeatable' — a good read on where working devs actually landed.

We need to re-learn what AI agent development tools are in 2026
— The tooling categories from 2024 don't describe what we're building anymore — this is a decent attempt at a new map.

The latest AI news we announced in June
— Google's own tour of everything it shipped in June — a useful skim to see how wide the front actually is.

From the big labs

Introducing Claude Sonnet 5 (Anthropic)
— Frontier agentic performance at $2/$10 per million tokens intro pricing — the price/performance line just moved again.

Redeploying Fable 5 (Anthropic)
— Back globally July 1 after the export-control pause, plus a cross-industry framework for scoring jailbreak severity.

Computer use built into Gemini 3.5 Flash (Google)
— Agents that see, click, and type across browser, mobile, and desktop — now native in the cheap fast model, not a research preview.

The weekly take

Sonnet 5 at $2/$10 is the real headline. Every time frontier-ish capability gets 5× cheaper, a class of app ideas flips from "unit economics don't work" to "weekend project" — and the tools above suggest plenty of people already noticed.

Have a great week — may your agents run green.

Issue #1 — Everyone's juggling coding agents now

Sun, 28 Jun 2026 00:00:00 GMT

First issue. This week's theme picked itself: everyone is suddenly running multiple coding agents, and the tooling to juggle them — switchers, shared knowledge graphs, plugins — is exploding faster than the agents themselves.

The tools

★ cc-switch (productivity, OSS)
Free desktop app that installs and switches between Claude Code, Codex, Gemini CLI, and Hermes Agent — one interface for all of them.
— 111K stars in a month is the tell — nobody runs just one coding AI anymore, and four terminal setups was getting silly.

Graphify (coding, OSS)
Builds a searchable knowledge graph from any codebase; works across Claude Code, Codex, Cursor, and Gemini CLI.
— Agents that query your project's structure instead of re-reading every file — stop paying the same context tax on every prompt.

HyperFrames (image-video, OSS)
Renders MP4 videos from HTML and CSS, built for AI agents to generate video programmatically.
— Video generation for people who'd rather write a stylesheet than wrangle a diffusion model — App Store previews, scripted.

Firecrawl Claude plugin (infra, OSS)
Firecrawl's web-scraping plugin with native Claude Code integration — 142K stars.
— Every agent eventually needs to read the web; this makes it a plugin install instead of a side project.

Worth reading

June 2026 AI Launch Wave: A Builder's Decision Map
— A sane map through a launch-heavy month — which of June's releases actually matter if you build things, and which are noise.

Production-Ready AI Agents: 5 Lessons from Refactoring a Monolith
— Google's engineers on why demo→production isn't a prompting problem — it's architecture, state, and guardrails. Scarred-knuckles material.

What Is Loop Engineering? Beyond Prompt Engineering
— The term everyone suddenly used this month — systems that prompt the agent, instead of you doing it. Worth knowing the vocabulary.

From the big labs

GPT-5.6 Sol previewed (OpenAI)
— System card landed June 26, preview June 28 — the naming (Sol, Terra, Luna) suggests a family, not a one-off.

Nano Banana 2 Lite (Google DeepMind)
— Their fastest, cheapest image model yet — the race at the bottom of the price curve matters more than the frontier for most apps.

Coefficient Bio acquired, AlphaFold's John Jumper hired (Anthropic)
— ~$400M all-stock plus a Nobel laureate — Anthropic is building a science org, not just a chatbot.

The weekly take

The picks-and-shovels layer is where the interesting stuff is happening. cc-switch hitting 111K stars in a month says the multi-agent workflow is already mainstream — the models compete, but the glue around them compounds. Watch the glue.

Have a good week — go ship something small.

AI Morsels

Issue #2 — Agents earn their keep

The tools

Worth reading

From the big labs

The weekly take

Issue #1 — Everyone's juggling coding agents now

The tools

Worth reading

From the big labs

The weekly take