Read time: 4 minutes

Welcome to the first real issue of Prime AI.

So here's the format.
Every week, in under 5 minutes:

- Latest: what happened in AI this week that actually matters to you.
- Steal This: a workflow, prompt, or template you can use immediately.
- Tool Spotlight: one tool I'm using, with an honest take.
- Signal / Noise: quick hits on what's worth watching and what to ignore.

Let's go.

Anthropic's head of product said something this week that stuck with me.
He called what's happening right now "vibe working" - the idea that you describe what you want in plain language and AI handles the execution.
Documents, analysis, research, presentations. Not just code.

The reason he said it?
Anthropic just released Claude Opus 4.6, their most capable model yet.
And having used it for the past week, I understand why.

The headline feature is a 1 million token context window.
In practical terms: I can now drop an entire client dataset, a full codebase, or a 200-page document set into Claude and it doesn't lose the thread.

I used to break things into chunks and hope it remembered part one by the time it got to part five. That's over.

Coding is noticeably sharper too. I've been testing it alongside OpenAI's Codex - both are strong. The gap between the top models is shrinking fast, which means the real differentiator isn't which model you pick. It's how well you use it.

My take: "Vibe working" isn't here for most knowledge work yet. But it's closer than people realise.
If you work with large documents or datasets, test Opus 4.6 this week. The difference is tangible.

This is the story that got everyone talking.
OpenAI started rolling out ads to free and Go tier users.
Ads appear below ChatGPT's responses, clearly labelled as sponsored.

The internet reacted exactly how you'd expect.
"ChatGPT is dead."
"Switching to Claude."
"The enshittification begins."

My honest take? Everyone needs to calm down.

Remember when Google introduced ads in search results?
People lost their minds. "Google is ruined." "Search is dead."

Fast forward to today and Google Ads is the backbone of the internet economy.
It funds the free product billions of people use daily. And it built one of the most powerful advertising platforms for businesses of all sizes.

OpenAI is burning cash. Massive amounts. The infrastructure to run ChatGPT for hundreds of millions of free users is extraordinarily expensive. Something had to give.

The details worth knowing:

Ads don't influence ChatGPT's actual responses = answers stay independent. Advertisers never see your conversations or personal data. No ads appear around sensitive topics. Users under 18 don't see them.

Free users can opt out.
The trade-off: fewer daily messages and reduced feature access (no image generation, no deep research).
The Go tier ($8/month) also gets ads, which has understandably annoyed people since they're already paying.
Plus ($20/month) and above remain ad-free.

Meanwhile, Anthropic ran a Super Bowl ad promising Claude stays ad-free.
Smart marketing. But it's also easy to say when you're not running a free product at OpenAI's scale.

What people are missing: This is a huge opportunity.
For businesses, ChatGPT advertising is a brand new channel where people are actively making decisions and exploring options. The businesses that figure this out first will have an advantage. Keep your eye on this space.

Here's something most people don't realise: your AI is agreeing with you too much.

Tell Claude or ChatGPT a terrible idea and watch what happens.
"That's an interesting approach!" "Great thinking!" It validates everything.
Which feels nice - until you realise your AI assistant is as useful as a friend who never tells you the truth.

The fix takes 30 seconds.

Go to Settings → User Preferences (Claude) or Custom Instructions (ChatGPT) and paste this:

The difference is immediate.
Instead of cheerful agreement, you get actual pushback. Honest feedback.
The kind of response you'd want from a sharp colleague, not a people-pleaser.

One more thing. When you're asking AI to review anything - code, a document, a strategy - don't say "review this."
Say "find problems in this, assume there are at least 3."
Forcing AI to hunt for issues instead of validating your approach completely changes the output.

Try both of these today. You'll never go back!

If you want to see a longer version, check the top comment on Reddit.

I've recommended a lot of AI tools.
Granola is the one where every single person comes back and says the same thing: "Why didn't I start using this sooner?"

It's a meeting transcription tool.
But calling it that undersells it.

Here's what makes it different: it's invisible to attendees. No bot joining your call.
No awkward "this meeting is being recorded by an AI" announcement.
It runs quietly in the background, merges your own notes with the full transcript, and produces structured action items and summaries.

I use it for every client meeting. Last week I had a 45-minute strategy call.
Granola gave me a structured summary with action items, owner names, and next steps - without me typing a single note during the call.
I shared it with the client ten minutes later.

It's free to start with and the free tier is generous enough to know whether it works for you.

If you take meetings - which is everyone reading this - try it this week!

Signal: Anthropic gave free Claude users access to file creation, connectors, and skills this week. If you've been using the free tier, it just got significantly more capable.

China's AI labs dropped several major models this week:
ByteDance's Seedance 2.0 (video generation) and Alibaba's RynnBrain (robotics). The gap between Chinese and Western AI is closing fast.

Noise: Every "Claude vs ChatGPT" hot take that treats this like a football rivalry. Use both. Use whichever works best for the task. Brand loyalty in AI tools is a waste of energy.

Predictions about AGI timelines - nobody knows. Focus on what works today!