| ISSUE 0x0A · TL;DR | 2026-09-21 · ~5 MIN |
Hello, builders 👋
Welcome back to your weekly dose of AI Bytes. I was relieved to see a small bit of agent housekeeping become news. Claude Code can now read AGENTS.md when a project has no CLAUDE.md: one fewer place for shared instructions to drift 📋
Today: Claude Code is making project instructions more portable, Mistral is bringing its models to Firefox Smart Window, and Alibaba’s open-code-review combines fixed checks with an LLM agent. The useful question is what each system lets us inspect and control.
|
My Favourite Picks
|
||||||||||||||||||||||||
📋 Claude Code reads AGENTS.md - Falls back when CLAUDE.md is absent; not yet available on Bedrock, Vertex or Foundry. Story of the week (Anthropic)
🗂️ Microsoft’s internal scraping alarm - Court filings reveal private concerns about training data acquisition. (TechCrunch)
🦊 Mistral comes to Firefox Smart Window - The browsing assistant beta adds Mistral models, with zero data retention agreed by the provider. (Mistral)
⚖️ OpenAI introduces Astra for Law - A legal research setup with a dedicated search index starts with selected law firms; API access is coming later. (OpenAI)
📷 OpenAI reportedly buys Glass Imaging - The reported $300 million deal brings in a camera startup founded by former Apple engineers. (TechCrunch)
✈️ FAA plans $875 million AI program - New software is intended to help air traffic controllers navigate their work. (TechCrunch)
"The more work we hand to agents, the more the instructions and checks matter."
| 0x02 | NEW AI PRODUCTS |
| 💳 CREEM 2.0 - Payments, tax handling, payouts and usage billing for software and digital products. (ProductHunt) | ▲ 575 |
| 🔑 tiun. - Authentication, billing, payments and customer analytics in one system. (ProductHunt) | ▲ 572 |
| 📬 Ami AI - Plans sales outreach from your website and targets, then runs approved campaigns. (ProductHunt) | ▲ 532 |
| 🎥 Naoma AI Demo Agent V2 - An AI demo agent that walks prospects through your product and logs sessions to your CRM. (ProductHunt) | ▲ 520 |
| 🎙️ Voiskey - Voice dictation that adjusts wording and tone for the app and audience. (ProductHunt) | ▲ 519 |
| 0x03 | TRENDING GITHUB PROJECTS |
| 🔎 alibaba/open-code-review - Combines deterministic checks with LLM review and line-level comments. | +15,028 ★ |
| 🧰 affaan-m/ECC - Skills, memory and security tooling for several coding-agent environments. | +6,265 ★ |
| 📝 ayghri/i-have-adhd - A skill that puts the answer first and keeps coding-agent output easier to scan. | +5,589 ★ |
| 🐋 stablyai/orca - Runs parallel coding agents across desktop, mobile and remote environments. | +5,404 ★ |
| 📚 Tencent/WeKnora - Turns documents into retrieval, a reasoning agent and a maintained wiki. | +4,867 ★ |
| 🌐 Panniantong/Agent-Reach - A CLI for agents to read and search content across social and developer platforms. | +3,914 ★ |
| 🛠️ addyosmani/agent-skills - Engineering skills intended for production work with coding agents. | +3,445 ★ |
| 0x04 | HN DEEP STORIES |
| 🎨 AI posters beyond the default look - A visual experiment in escaping the default AI poster aesthetic; a June post resurfacing on HN. (HackerNews) | 1623 pts |
| ⏳ Everyone should slow down AI except me - A critique of the double standard in calls to slow AI development. (HackerNews) | 811 pts |
| ✍️ How to write with an LLM - Use the model to flag weak prose while keeping the wording and editorial judgment yours. (HackerNews) | 678 pts |
| 🧠 Why agents lie, cheat and coordinate - Yoshua Bengio examines how reward-seeking and imitation may produce deceptive agent behavior. (HackerNews) | 657 pts |
| 🧮 Bend puts proofs beside AI-written code - A language using proofs to check declared properties, with native and GPU execution. (HackerNews) | 606 pts |