Welcome back. AI is narrowing the competitive gap for many companies — leaving founders to wonder what's next. Snap is betting on hardware, they having recently released the standalone AR glasses for $2,195. The market reacted swiftly, sending the stock down by 10% but the CEO's rationale behind the move is worth watching.

Also: OpenAI engineer shares 3 ways Codex can use a computer, a skill that judges your web animations, and a startup founder reveals which department uses Claude most.

Anthropic turns Claude Code sessions into live shareable pages: The AI lab just rolled out artifacts, which turn your coding sessions into live web pages using your full context, including the codebase, connectors, and conversation. It’s perfect for PR walkthroughs, dashboards, or incident timelines that update in real time while Claude works. This lets the team skip the status update and see the same view. Everything stays private to your org, and it's currently in beta for Team and Enterprise plans.

OpenAI brings recorded Mac workflows to reusable skills: The ChatGPT maker just shipped Record & Replay for Mac. It watches you do a repetitive task once, like filing an expense, creating a configured issue, or pulling a recurring report and turns it into an editable skill. Just start a new thread, give Codex the new inputs, and it'll handle the rest using Computer Use, browser actions, and plugins. Watch how it works.

Cursor ships a skill to set up automations: The AI coding startup just dropped /automate, which turns a plain-English description of a repetitive task into a configured run. It handles wiring up the triggers, instructions, and tools for you. Teams can fire automations from a Slack emoji reaction. You can also use new GitHub triggers for issues, reviews, and finished workflow runs. Cloud agents kicked off this way can now use their own computer to produce a demo of the work.

Anyone can make an AI demo. The hard part is shipping it, monitoring it, and knowing when it breaks (and how to fix it).

Datadog's Developer Toolkit for the AI Era gives you the playbook for building, deploying, and observing AI applications at scale. It covers: CI pipelines, LLM observability, feature flags, testing, and modern AI delivery workflows.

Get the toolkit and ship with confidence.

Agents stopped working solo. Multi-agent systems have become the standard in production, with orchestrator-worker setups accounting for roughly 70% of deployments. In these systems, one agent spawns others and hands off tasks to run in parallel. While the work gets faster, the coordination definitely gets trickier.

This breaks the usual playbook. The standard fix is to skip coordination altogether. You give each sub-agent its own context window. That way, they never see what the others are doing. This works fine for independent tasks. But everything falls apart the second one agent changes an interface that the others are relying on. When that happens, the pieces just don't fit.

And the cause is almost never the model. A recent study of over 1,600 agent traces tied 37% of failures to coordination and 42% to weak specs, not reasoning. Context usually gets lost during the handoff. The lead agent ends up waiting on a report that fails to frame the actual task. If you run it ten thousand times, those same breakdowns stay hidden deep in the traces.

The recipe. Anthropic just dropped a cookbook that strips multi-agent systems down to the parts that usually break. It ignores the specific tasks and domain logic to focus entirely on messaging and lifecycle mechanics. By using a shared hub for direct messaging and real-time tracking, it eliminates polling and lost data. It's a clear sign the real work has moved from model intelligence to the plumbing between agents.

AI shows up in 60% of engineering work. Only about a fifth of it can be handed off without someone babysitting the output. That’s because agents are still missing the context you have. Join live June 24 (FREE) to find out how teams pulling ahead are using a context layer to level up.

Register Now

Messy code usually sneaks in during a rush, and waiting for a PR review is often too late. You can use OpenAI's non-interactive mode to run Codex as a git pre-commit hook that checks staged changes and blocks the commit if it hits a critical issue.

To set it up, run "touch .git/hooks/pre-commit && chmod +x .git/hooks/pre-commit" and paste a headless Codex review script inside. With read-only mode enabled, Codex will flag risks in your diff without modifying your files.

Now, git commit runs the review automatically. A clean diff commits like usual, but any critical bugs stop the commit before it ever hits CI or a peer review. You can also drop that same codex line into a GitHub Action to keep the same guardrails in your pipeline.

P.S. Get 50+ AI coding hacks for Claude Code, Cursor, and Codex here.

Loop engineering in Claude Code + Codex: You'll learn how to build autonomous AI workflows using Claude Code and Codex. This tutorial shows you how to replace manual prompting with automated loops, such as scheduled tasks and goal-driven sub-agents. You will also pick up practical tips for managing token costs and ensuring precise execution.

Cloudback: This tool runs automated, snapshot-based backups for Linear workspaces, GitHub, GitLab, and Azure DevOps. It makes sure your code, project history, and important metadata are securely backed up and easy to recover.

Flue (5.6K ⭐): A TypeScript framework that lets you build AI agents using the same harness-driven architecture as Claude Code. It allows for truly autonomous software that you can write once and run anywhere, from local CI to the cloud.

Trigger a Workspace Agent from the API (by OpenAI): Saved ChatGPT workflows typically need a manual push, which makes them tough to trigger from outside systems. By setting up an asynchronous API trigger, you can let those external systems kick off agents automatically and send the results straight to your connected apps.

This guide explains how to set up loops that keep your agents working around the clock (1.5M views).

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 300K+ 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