In the days leading up to Super Bowl weekend, Anthropic and OpenAI dropped their latest coding models, with Anthropic releasing Opus 4.6 and OpenAI shipping ChatGPT-5.3-Codex.

Just ahead of that, OpenAI also released a macOS desktop app for Codex and doled out 2x usage through April when using it.

OpenAI also seems to be looking to cash in on some turmoil following Anthropic’s crackdown on subscription usage in tools like OpenCode by leaning into allowing their subscriptions to be used with those tools.

The extra usage and openness around editor support may be a sign that OpenAI is playing a bit of catch-up on the coding front following their code-red triggered by Google’s Gemini 3 release.

We’re even seeing the AI battlefield spill into mainstream culture. Anthropic ran Super Bowl ads taking shots at OpenAI’s upcoming inclusion of ads, while also poking fun at ChatGPT’s perceived sycophancy. Though that may have been harder to pick up on if you’re not working with these tools every day.

Still, not quite as groan-inducing at my watch party as Coinbase pumping people up with Backstreet Boys karaoke, only to rug-pull us all with the Coinbase logo at the end.

It’s a good reminder that while AI agents and agentic coding tools dominate conversations in technology circles, they haven’t fully broken into the broader public consciousness yet. Those of us on the bleeding edge may be using these tools to build personal software and automate workflows, but those ideas are still fringe outside our bubble.

We’re in a moment where the landscape is shifting quickly. The most useful thing we can be doing right now is experimenting and learning what these tools are good at, where they fall down, and how to apply them responsibly as they move toward wider adoption.

🧠 New Model
Introducing Claude Opus 4.6
Opus 4.6 builds on the momentum of Opus 4.5 with a model that plans more, runs longer, and does a better job checking its own work. It also introduces a beta 1M token context window, which is a big jump from the 200K window in 4.5. What I’m most curious to see is whether this release eventually unlocks the agent swarm feature that was found buried behind a feature flag.

🧠 New Model
Introducing GPT-5.3-Codex
Codex has been picking up noticeable steam in my feed lately. What once felt overwhelmingly Opus 4.5 dominated has shifted toward more discussion and support of Codex. OpenAI may be playing the extra usage card at just the right time, pairing it with a new model release to build momentum and take a swing at Claude’s crown.

💻 New App
Introducing the Codex app
A few days before GPT-5.3-Codex dropped, OpenAI released a Codex desktop app for macOS. It adds a bit more GUI polish to replace the Codex CLI TUI while keeping a similar feature set, most notably the ability to manage multiple agents in parallel. I’ve only spent a couple of hours with it so far, but I appreciated that it carried over the skills I had already set up in the Codex CLI.

📰 Article
Unrolling the Codex agent loop
We might as well keep the Codex train rolling. This article breaks down the Codex agent loop and does a solid job explaining what agentic harnesses are doing under the hood, coordinating user input, model output, context, and tool calls into predictable workflows.

Podcast
Pi – The AI Harness That Powers OpenClaw w/ Armin Ronacher & Mario Zechner
Syntax.fm had Armin Ronacher and Mario Zechner on to talk about Pi, the agent harness behind tools like OpenClaw. They cover a wide range of agentic development topics, including why Bash is “all you need,” how workflows are changing, and where agentic AI tools seem to be headed.

Thanks for reading.