after every second brain I've tried, this one stuck

📌 TL;DR

• Claude Opus 5.5 → the best model I've ever used

• OpenAI's new models → GPT-6.1 Sol gets close to Astra for a fifth of the price

• OpenAI Dots → an always-on agent with its own cloud computer.

• DevDay → ChatGPT is turning into an operating system

• Grok 4.7 → holy sh*t thats a lot of new models

• Builder's notes → under the hood of my ‘second brain’, plus a whole business, brand and live website, from idea to launch in about 3.5 hours.

Claude Opus 5.5 is the best model I've ever used

Anthropic released Opus 5.5 on 22 September. Hands down, it's the best AI model I've ever used.

It's my new daily driver for everything I do, and it's incredible at design.

With models now, it's all about vibes, not benchmarks. Opus 5 was one of the highest-scoring models ever on paper, yet the internet collectively agreed it was the worst AI model anyone had ever spoken to.

Opus 5.5 fixed that.

What actually changed

It's nicer to work with. It takes feedback without arguing, and its writing is far easier to read. It puts the important thing first, where Opus 5 would walk you through every step before getting to the point.

It's cheaper and faster: 20% cheaper than Opus 5 and less than half the price of Fable, Anthropic's top model. It also answers more than 30% faster.

Where it shines, and where it doesn't

For planning and analysis, it's as good a thinking partner as Fable. But for coding, it gets roughly 90% of the way there.

Opus 5.5 is the first Opus that's as easy to work with as it is smart, and that's why so many people have moved their daily work back to Claude.

So if you only use one model right now, make it Opus 5.5. It's cheaper than the last one, it's great at almost everything, and you'll only need Fable for the very hardest problems.

And Sonnet 5.5, the cheaper sibling

Sonnet 5.5 followed on 28 September. Think of it as the everyday model: not as deep a thinker as Opus, but quicker and much cheaper to run.

It costs the same as Sonnet 5 and half as much as Opus 5.5. Anthropic says it's over 30% faster, and because it gets jobs done in fewer steps, a typical task costs up to 30% less.

Where it fits….

The general feeling is that it's a clear step up from Sonnet 5 at low and medium effort, and great for brainstorming, iterating on ideas and design work. But if a task needs it to think really hard, don't crank Sonnet’s thinking level up. Move to Opus instead.

I haven't used it much myself, but Sonnet is what I always reach for when I'm running scheduled tasks in Claude that don't need a lot of effort, like my morning brief.

OpenAI's new models: GPT-6 Sol, Luna and 6.1 Sol

OpenAI filled out its GPT-6 family, the same day Opus 5.5 landed.

please bare with me on all these numbers 🤓

OpenAI pitches 6.1 Sol as close to Astra for about a fifth of the price.

You can use it in Codex (OpenAI's coding agent) and ChatGPT Work, but not in regular ChatGPT chat yet.

I haven't spent much time with these models personally. But from what I've seen online, and my limited time using them, the general consensus is that Sol isn't quite Astra, but it's close enough for everyday work and noticeably faster.

GPT sol also excels at writing work (easily my fav writing model)

Every tested an early version of Sol on computer use (the AI clicking around software for you), and it matched Astra on 17 of 18 attempts for a fraction of the cost.

On Every's knowledge-work scoreboard, GPT-6.1 Sol came out on top at 86%, ahead of Opus 5.5 at 80%.

Where it falls short is long, hands-off coding builds, and design work, where Opus 5.5 still takes the crown.

So the rough split: If you live in Codex for reading, writing and getting things done, Sol fits. For hard coding and visual projects, reach for Opus.

OpenAI Dots: one more always-on agent

OpenAI launched Dots at DevDay on 29 September.

A Dot is an AI agent with its own computer in the cloud that's always on. You assign it work, and it keeps going even with your laptop shut.

You can reach it from your phone, the ChatGPT app, Slack or Microsoft Teams.

Dots are completely personal, though. You cant share dots with your team (which defeats the point of them imo)

Dots vs Grok Bot vs Muse

If this sounds familiar, it's because Dots, Grok Bot and Muse are pretty much selling the same thing: an always-on assistant with its own cloud computer that connects to your apps and keeps working whether your laptop is on or not.

As Chase breaks down in his video, the real differences come down to three things.

Power. Dots runs on GPT-6 Astra, which is comfortably the strongest of the three. Grok Bot runs on Grok 4.7, and Muse runs on Meta's Spark 1.3, the lightest model of the lot.

Price. Muse has a free tier, and Grok Bot is $20 a month. Dots needs a Pro plan, which starts at $100 a month, so it's a big upfront cost.

What you actually need it for…

If you mostly want help with everyday stuff, like checking your emails and asking quick questions, Muse is the easiest and cheapest place to start.

If you want an assistant that directs other agents and builds things on its own computer, that's where Dots and all that extra power earn their price.

Grok Bot sits somewhere in between, and it's a natural fit if you already use Grok or Cursor.

I personally don’t use any (just Instinct as my PA)

We're in another awkward transition, like when the world went from chatbots to agents.

This time it's from agents on your own computer to agents in the cloud that your team shares, and nobody has the split figured out yet.

I started using Grok Bot a lot, then quickly stopped, because it couldn't do anything I wasn't already doing with Claude or Codex.

So be ruthless with your time when it comes to agents, and remember that AI is not the goal.

If you find yourself spending four hours a day tinkering & migrating from Claude to Grok Bot, and then to dots, and moving over skills and instruction files, slap yourself in the face and wake up. It's a complete illusion of productivity.

Grok 4.7

xAI (now calling itself SpaceXAI) released Grok 4.7.

It's built for long, multi-hour tasks and checks its own work more.

The price hasn't changed, but it uses about twice as much per task, so each job ends up costing more.

I haven't used it much personally, but Kun Chen put together a great write-up after a full day using it as his main agent.

His verdict: "a really solid model with visible improvements over 4.5". It sticks closely to your instructions, asks before acting, and is predictable, with no genius moments but no nasty surprises either.

The trade-off is that it's slower and pricier than 4.5.

Also this week...

• Google announced Gemini 4 Argon, but you can't use it yet → Google's new flagship model. For now only vetted cyber-security teams get it, with paying developers and Google AI Ultra subscribers next, "as soon as possible".

• Claude Code won't cut out mid-task anymore → when you hit your five-hour usage limit, Claude Code (Anthropic's coding agent) now borrows a little from your weekly allowance to finish what it's doing cleanly, instead of stopping halfway through. a nice touch.

• Grok Team Bots are now in public beta → Now teams on Grok's Teams and Enterprise plans can set up shared AI assistants, each with its own role, skills and app logins, and talk to them in Slack. Everyone's chats with the bot stay private.

• Manus 2.0 and Cue → Manus rebuilt its agent. It also launched Cue, their answer to grokbot, dots, muse, etc

💡 Builder's notes

inside the most effective second brain I’ve ever built

I've been using mymind for coming up on 12 months. It's essentially a bookmarks bar for the entire internet.

Highlights from emails, any web page, any Instagram post, any Amazon listing, you name it. If it's on the internet, you can save it into mymind.

The problem was my agents couldn't get to any of it. There was no MCP (the connector that lets an agent like Claude search and use another app).

Well, that has now changed.

Saving things into mymind and having my agents find them is now one of the best workflows in my life.

A lot of people try to build this with Obsidian, a web clipper and a pile of bootleg solutions. Even I tried.

You end up tinkering, fixing it when it breaks, and setting it all up again. By the time you're done, it's eaten the time it was meant to save you.

I can confidently say mymind is the best solution for this.

A whole business, from idea to launch, in about 3.5 hours

This week I recorded two new masterclasses for The AI Course community…

1. Vibe branding masterclass

2. Landing page/ funnel building masterclass

To film them, I made up a business from scratch. Tenfold, an agency that places AI-native virtual assistants.

The branding took me 30 minutes. Logo, colour palette, type system and a full brand sheet.

The website took about 3 hours. It's live at tenfoldva.site, with a low-ticket offer funnel and a book-a-call funnel, fully wired up to my calendar through Cal.com and to Stripe for payments. (Tenfold is made up, so don't go booking a VA.)

That's a full business idea to launch in about 3.5 hours.

still so wild to me that you can do that.

How I built it

Claude Opus 5.5 did the design work, with Paper as the canvas. Paper connects to Claude through its MCP, so Claude can see and edit the designs directly.

The process:

• collect inspiration (logoinspo.com and things I'd saved in mymind)

• give Claude the references

• brain-dump the brief out loud with Wispr Flow instead of typing it

• review round one, lock the logo mark, pick the font

• export the brand sheet

It's as if you have a graphic designer in front of you and you have unlimited revisions.

It's truly insane how quickly you can move if you know how to use AI properly.

Both are recorded raw and unscripted inside The AI Course.

slowly turning into a paper fan page, but I had to include this one…

I used Paper to make a pop-up graphic title for one of my Instagram Reels this week. It's sitting at over 300,000 views.

I'd seen a guy on Instagram who does a Free Font Friday series and loved the way his graphic pops up on screen. So I made my own version for a new series: Skill of the Week.

🧰 Tools to try

• Space → A cloud drive your agents can read and write to. It shows up as a normal drive on your computer, files stream in only when you open them (so they never fill up your laptop), and every change is versioned. Put your project folders in Space and you and your agents work off the same files.

🥣 Brain food

• how to think clearly → Genuinely one of the most life-changing pieces of content I've ever stumbled across. I've watched it three times and I'm rereading a written version on my Kindle every night before bed. 1,000,000% worth your time.

• never write AI slop again → If this is an "eat your vegetables" episode, these veggies were sautéed in butter. Writing is in everything I do, and this one changed how I work well beyond writing.

• Clay's AI Writing Policy → I love this idea, and think every company needs one of these

• Where to put your AI context → local, notion, or github?

• The SEO and AEO setup I run with AI agents → Open the article, Cmd A, Open Claude Code, Cmd V, Enter, Win.