Welcome back. Working with AI still feels like a one-way street: you think, and it responds. True collaboration needs an interface that listens, sees, and reacts in real time without constant prompting. The AI lab founded by OpenAI's former CTO Mira Murati just unveiled a model that does exactly that.

Also: Find and fix bugs 10x faster with this new Codex workflow, set up Hermes Agent (the rising OpenClaw rival) in 30 minutes, and make sense of the tech job market paradox.

Thinking Machines previews its first interaction model: Former OpenAI CTO Mira Murati’s AI lab just released a research preview of TML-Interaction-Small. It’s a 276B-parameter model that processes audio, video, and text as continuous 200ms streams rather than separate turns. This allows the model to interrupt, provide backchannel feedback, and react to visual cues in real-time before you even finish speaking or typing, delivering responses in just 0.4 seconds. See it in action.

OpenAI unveils a frontier AI platform for cyber defenders: The ChatGPT maker just unveiled Daybreak, which pairs GPT-5.5 models with Codex Security to find vulnerabilities and generate patches inside code repositories. It builds an editable threat model from each codebase, narrows analysis to realistic attack paths, and then verifies every fix. They also launched a $4B Deployment Company, after acquiring a consulting firm Tomoro, to embed 150+ engineers directly into businesses to build out production-ready AI.

Anthropic ships unified view for running coding agents in parallel: The AI lab just rolled out Agent View in Claude Code, a research preview that brings all your active sessions into a single list. This finally puts an end to the constant tab-swapping typical of parallel workflows by showing you exactly which agents are finished, still working, or waiting for your input. You can peek at any session to reply inline or move tasks to the background using the /bg command.

Your dev stack got an AI upgrade everywhere except the input layer. You're still typing every prompt, every ticket, every review comment by hand.

Wispr Flow closes that gap. Dictate into Cursor, VS Code, Slack, Linear, or anywhere else you work. It's syntax-aware: camelCase, snake_case, acronyms, and file names all come through clean. Mention a file in Cursor or Windsurf, and it auto-tags.

It's the voice layer for an AI-native workflow. Speak your intent. Your tools do the rest.

Available on Mac, Windows, iPhone, and Android. Used by millions of developers, including teams at OpenAI and Mercury.

Try free

Shopify is breaking away from the AI status quo. Most AI coding tools operate behind closed doors. Cursor sits between you and your IDE, while ChatGPT and Claude live in private tabs that nobody else sees. But a few days ago, Shopify CEO Tobi Lütke revealed that their internal coding agent, River, does the exact opposite. River skips the Slack DMs and works entirely in public Slack channels. It’s a move that's paying off. River now handles one out of every eight PRs merged into their monorepo weekly.

The whole company is watching. River reads code, runs tests, queries the data warehouse, and opens PRs. Lütke describes this setup as a Lehrwerkstatt, a German tradition where apprentices learn simply by being near the masters.

Working in public is paying off. In just two months, River’s merge rate shot up from 36% to 77%. This wasn't because of a model upgrade; it happened because engineers watched River mess up in public and corrected it in real-time. Linear CEO Karri Saarinen says using their agent feels like "accessing the company brain." It opened 1,330 PRs last month alone, autonomously resolving 30% of bugs.

Shopify cracked the playbook. Django co-creator Simon Willison notes that Midjourney did the same on Discord, teaching people how to prompt by letting them watch one another. Keeping AI conversations private means leaving leverage on the table. When you make the work visible, every prompt becomes a lesson for anyone watching, and that’s the new AI playbook for companies.

Renaming a function across 200 files or swapping “axios” for “fetch” everywhere usually eats up an entire afternoon. The “/batch” command fixes this by splitting the workload across parallel agents:

It breaks the task into 5 to 30 units, presents a plan, and then runs each unit in its own “git worktree” before opening one PR per agent.

P.S. You can find 50+ AI coding hacks here.

How to turn Karpathy's LLM knowledge base into a production agent: This tutorial shows developers how to build an automated knowledge base inspired by Andrej Karpathy’s approach to compiling and organizing personal notes. You’ll learn how to use Claude Code and Obsidian to build an agent that automatically turns raw web clippings into structured, interconnected atomic notes.

PRFlow: AI code reviewer that catches what others miss. Think of it as an automated teammate that reviews every pull request before it ever hits production. In tests across 10 real-world projects, PRFlow identified seven critical security vulnerabilities that other tools completely overlooked.

Awesome Hermes Agent (2.8K ⭐): This repo is a curated toolkit for the Hermes Agent ecosystem, bringing together the best skills, plugins, and deployment templates. It provides the essential building blocks (from memory backends to multi-agent setups) to help you build and scale a professional Hermes-based agent stack.

Build iterative repair loops with Codex (by OpenAI): Technical documentation often suffers from broken or outdated code examples. Implementing a closed-loop AI workflow that reviews, repairs, and validates the code solves this by improving the output using automated feedback.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 270K+ engineers and 150K+ followers on socials. Get in touch.

You can also reply directly to this email if you have suggestions, feedback, or questions.

Until next time — The Code team