SUBSCRIBE | ADVERTISE

Opus 4.6 = feel the AGI

Essential AI news from the last week + my initial impressions of Claude Opus 4.6

Was this email forwarded to you? Sign up here.

Howdy wizards,

News first, then my impressions of Claude’s newest (and frankly brilliant) Opus model.

Here’s what’s brewing in AI.

Models

• Anthropic released Claude Opus 4.6 with a 1M token context window, and agent teams in Claude Code. It’s currently the world’s top rated model for all things text and code on Arena. You can read some of my early impressions later on in this newsletter. I am in awe.

• OpenAI released GPT-5.3-Codex, expanding Codex into a full computer-use agent with state-of-the-art coding skills. Something notable is how both Anthropic and OpenAI are actively saying that their previous model helped build the new ones ie. “We build Claude with Claude” and “we used Codex to train and deploy GPT‑5.3-Codex”. I wouldn’t get too distracted by who’s running ads and who is not (more on that in a bit). This sounds like an early taste of the recursive self-improvement promise of AI, and might be the frontier most deserving of your attention right now.

• OpenAI also launched the Codex desktop app for macOS. It allows users can run many agents in parallel. The dedicated tab for automations, which lets you set up a workflow and run it on a schedule automatically, might be powerful and quite accessible to less-technical people as well.

• The OpenClaw (previously Clawdbot/Moltbot) virality is still going strong. ICYMI, it’s an open-source AI agent that autonomously controls computers (it got 150K+ GitHub stars in 72 hours). Has triggered some security warnings from serious people, especially in terms of prompt injections; it doesn’t seem to be exaggerated given the level of access to your accounts + autonomy that the agent has.

• OpenAI launched Frontier, a platform for deploying governed AI agents across enterprise systems with permissions, identities, and auditability. Anthropic and OAI are both working very hard to gain traction in the enterprise segment -- whoever owns the agent orchestration layer there will cash in massively.

Industry moves

• Elon Musk’s SpaceX acquired xAI, creating the world's most valuable private company at $1.25 trillion valuation. xAI was losing tons of money while SpaceX was very profitable, so now Musk can use SpaceX cash to fund his otherworldly AI expansion : "within 2-3 years, lowest cost AI compute will be in space.", says Musk.

• Anthropic ran an $8M Super Bowl ad mocking OpenAI's decision to put ads in ChatGPT. Sam Altman responded with a disappointing lack of humour, calling it "dishonest." The ads feature 30-second satirical skits of AI assistants interrupting users with sponsored content. The brands got the attitude of early Apple marketing and I’m not gonna lie, that simply sits right with me.

Research

• NASA's Perseverance rover completed the first AI-planned drive on Mars. The rover drove 500 meters, planned entirely by Claude. In case there was any doubt—it really does more than just summarising emails.

• METR reports GPT-5.2 can autonomously complete software engineering tasks that take humans 6.6 hours now, up from 4 minutes for GPT-4 two years ago. It’s doubling every 4-7 months.

IN PARTNERSHIP WITH GRANOLA

Turn your meetings into momentum

Meetings aren’t the problem. It’s everything that comes after them.

The context switching, the cognitive load of remembering what you promised, the fear of something important slipping through…

Granola lets you take notes the way you always have. It works in the background, turning conversations into clear summaries, action items, and next steps.

With Granola, your meetings are no longer about scrambling to keep up. They lead somewhere.

Download Granola for free and use it in your next meeting using the code WHATPLUGIN →

❦

What I’m actually using

• Testing Claude Opus 4.6 has been a total feel the AGI experience for me. Here’s what I’ve been doing with it this week:

  • Coding: What I’ve noticed so far from developing in Claude Code is that the model goes on longer (often 15 min+) coding-runs and comes back with working, excellent results.

  • Creative/writing: It’s also an awesome sparring partner for creative ideas if it has enough context; I’ve implemented several features I’m really happy with in the last week that came directly from brainstorming with Opus 4.6. It also seems to have become meaningfully better on writing, I’ve been doing lots of copywriting this week and it reasons exceptionally well about word choices. It still needs informed decision making, system design, creative direction and taste to make great things, but the technical execution aspect is pretty much handled.

  • Data analysis: I’ve had Opus 4.6 create some data visualisations that might have taken me days to code up on my own in something like R or Python. As long as you start off from a clean, tidy dataset (an essential pre-requisite) you’ll be able to create any data viz you want in a matter of minutes.

    

    I’ve been working on gathering AI implementations and breaking them down into a structured dataset. Visualising the data with Opus 4.6 has been a delightful experience.

     

• I’m also enjoying Claude Code’s new feature where it delegates the work to different agents. It feels like I’m communicating with a middle manager with excellent oversight, who delegates the work to agents who actually carry out the work. Claude Code now even has a dedicated Agent team feature, where you can specify teammates with greater control of the delegated work, including talking directly to different team members; I haven’t tested Agent team yet, as it seems like it would be a bit of over-structuring to me, but the default Claude Code experience now has elements of the same thing: simple, automatic delegation that works great.

  

  Agent speed is definitely not the bottleneck anymore—it’s us.

• I'm currently burning tokens like we burn firewood up here in Trondheim (which is to say, aggressively). So I’ve had to upgrade my Claude Max subscription again from 5x to 20x ($200/mo) last week. I gotta admit, that’s a substantial amount for a subscription, but we’re in the most exciting time in history to make something wonderful, and I don’t want my flow to get interrupted by rate-limits.

Receiving feedback from you in the poll below is like the first cup of coffee in the morning for me.

THAT’S ALL FOR THIS WEEK!

Disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes work with sponsors and may earn a commission if you buy something through a link in here. If you choose to click, subscribe, or buy through any of them, THANK YOU – it will make it possible for me to continue to do this.