Every few weeks there's a new AI model, a new chart full of benchmark scores, and a new round of people arguing about whether we've finally reached AGI. It's a fun debate to watch. It also has very little to do with how my week actually goes.
Here's my take. Claude Opus 5.5 isn't AGI. When I started writing this, I assumed Anthropic's own Fable and OpenAI's Astra were a level above it. They're not. The independent leaderboard actually puts Opus 5.5 on top.
And still, it's the model that broke my productivity.
Source: Artificial Analysis Intelligence Index, October 2026, each model at its highest effort setting.
One model for pretty much everything
The first reason is range. I use the same model for things that used to need completely different people.
It finds bugs and fixes them in minutes. A few days ago, clicking anywhere inside my website's article editor was secretly pressing the Bold button. Claude tracked down why, fixed it at the source, and tested it, faster than I could have explained the problem to a developer.
It builds apps. I rebuilt my personal website from scratch in one evening and a morning. I revamped Nocta, my product shot app, including a bug where exported images didn't match the preview. Since then, I've built 2 other apps and have 3 more in the pipeline. All in the span of a few weeks.
Nocta, my product shot app, after the revamp.
It makes videos. Launch videos, product demos and ads, without me opening video software.
A frame from the mktbox launch video, made entirely with Claude Code.
It designs. Nocta's new logo, the covers for these articles, the visual style of everything I put out. And it doesn't do it mildly either. I'm talking about vectorised logos and high quality covers, something that would require a designer and Figma not long ago to do a proper job.
An article on this site, with a cover from the cover maker Claude built into it.
And it handles the everyday stuff, for my own projects and for my day job as Head of Marketing at LayerX. I'm pretty sure there's a lot more it can do that I haven't even tried yet.
It gets what you mean from a rough start
The second reason surprised me the most.
I'm building a new app right now, and it has some real complexity to it. What the user does first, what's free, when they pay, what they get at the end, what they can still change afterwards. I had a rough idea and a bunch of open questions.
I explained it the way I'd explain it to a friend over coffee. Messy, half-formed, jumping between ideas. Claude asked a few questions, then laid out the whole flow and the features, and it was exactly where I wanted to go.
No other model I'd tested before got there this fast and this accurately. That's the part that changes things for someone like me. I'm a marketer, I don't write specs, and I don't need to anymore.
The real cost is per output
The third reason is price, and this is where it gets a bit silly.
I'm on the Claude Pro plan. It has short limits that reset a few times a day, and a weekly limit. I hit the short ones pretty much every day, and at most I wait an hour or two for it to reset. The weekly limit is the one I actually have to manage. As I write this, I'm past 60% of it, and it only resets on Thursday.
But look at what those $20 of budget buys. A full video, from the first message to the final file, takes about 10 to 20% of one daily limit. I can make four videos in a row before I hit it. Building an app takes around 20 to 30% of my weekly limit, depending on how complex it is.
What I make | Share of my Pro plan limits |
|---|---|
One video, start to finish | 10 to 20% of a daily limit |
One app | 20 to 30% of the weekly limit |
Four production-quality videos in a row, on a regular subscription. A year ago that was a $10k agency invoice.
Why I'm not chasing other models
I haven't tried Fable or Astra, and I don't feel the need to. From what I've seen, Fable needs extra credits on top of the plan and costs more per task. Astra is cheaper per task, but it scores lower and would mean paying for a second subscription.
Unless you're building something truly wild, or you have the budget to burn, I don't think it's worth it. I don't need the best model in the world. I need to know what my model delivers and where it could improve. A smarter model that makes every video cost five times more doesn't make my work five times better.
Where it could be better
It's not perfect. When I started making videos, every small tweak, like moving a subtitle a few pixels, meant a full new render. I was the one who suggested building an editor so I could make those changes myself. Claude built it, and built it well.
But if it had thought of that on its own, it would have saved me a lot of time and tokens. It's great at doing what you ask. It's less great at noticing what you should be asking for.
What this means for you
If you're waiting for AGI before you take AI seriously, you're waiting for the wrong thing. The shift is already here, and it comes from one model that's good enough at a huge range of work.
Take a bow, Opus 5.5.