Hey {{first_name | there}},

Yes, Spotify's CEO just admitted that their top devs haven't written a single line of code since December. 

I'll get to the reason why in a bit, but there's a lot more going on this week that you should know about.

17,000 people have pledged to cancel their chatGPT subscriptions and 700,000 have signed up to spread the word over OpenAI's political ties.

At the same time, there's a free model that's matching Claude and GPT on real benchmarks. (China is never behind)

And Anthropic dropped a 32-page guide on building with Claude that's actually worth your time if you use it every day.

Before we get into it, a quick question:

Here's what happened: OpenAI President Greg Brockman and his wife donated $25 million to a pro-Trump super PAC. 

Separately, US Immigration and Customs Enforcement uses a resume-screening tool built on GPT-4. 

Put those two facts together, and you get QuitGPT, a boycott campaign that's attracted 700,000 sign-ups and gone viral enough to pull in actor Mark Ruffalo.

The campaign urges users to cancel their ChatGPT subscriptions and spread the word. 

Sam Altman hasn't addressed it. That silence is a choice.

Marketing professor Scott Galloway, whose video apparently inspired the movement, made a simple argument: hit them in the wallet. Make cancellations loud enough to become a liability.

Whether it works depends on one question: Can 700,000 sign-ups move the needle when ChatGPT has 800 million users and only 5% pay?

The QuitGPT movement gives you a reason to pause and actually audit this. Not for political reasons, just practically. When did you last seriously try an alternative? This week is a good time.

A Chinese lab called Z.ai dropped GLM-5.

It’s a 744 billion parameter open-weight model, meaning the weights are public. Anyone can download and build on it, and it's currently sitting at the top of the Intelligence Index.

On coding benchmarks, long-horizon reasoning, and a simulated year-long business challenge called Vending Bench 2, it's trading blows with closed models like Claude and GPT.

The gap between open and closed models used to be 12+ months. Now it's measured in weeks. 

And unlike proprietary models, open weights get better faster, thousands of developers improve them for free, bugs get fixed in days, new capabilities appear that the original teams never built.

During Spotify's earnings call, co-CEO Gustav Söderström told us that the company's best engineers haven't written a single line of code manually since December.

They're using an internal AI system called Honk, built on Anthropic's Claude Code. 

The workflow: engineer sends instructions over Slack, sometimes from their phone before reaching the office, Claude writes the code, the update gets deployed, engineer reviews and merges it. All before the morning standup.

Result: 50+ new features shipped in 2025. Dramatically faster testing, deployment, and product velocity.

Anthropic released a guide on building “Skills” for Claude.

A Skill is a reusable instruction file (SKILL.md) that gives Claude consistent behavior across sessions. Instead of rewriting prompts every time, you define how you want a task handled once, and Claude keeps using it.

That’s the vision.

In reality, it’s still early. Claude sometimes forgets Skills are persistent, overwrites parts of them, or replaces full files with summaries. Developers are using workarounds like re-uploading full files to avoid losing context.

So this isn’t a finished system.
It’s a signal of where AI tools are heading.

We’re moving from prompting to structured AI workflows. Less “ask a question,” more “build a repeatable system.”

Here’s how you can go through this entire guide in less than a minute.

If you step back and look at everything that happened, one thing becomes clear:

The AI model itself is no longer the advantage.

Models are improving everywhere. Open-source versions are spreading fast. Access is getting cheaper and easier. Having a “better model” isn’t enough anymore.

Spotify understands this. They’re not just investing in more engineers, they’re using the behavioral data they already own.

GLM-5 shows how quickly models are becoming commodities. When open weights are available, anyone can build on top of them.

The QuitGPT backlash highlights something even more important: OpenAI’s biggest risk isn’t a stronger competitor. It’s losing user trust. And trust can’t be fixed by increasing parameters.

At the same time, AI is shifting from answering questions to actually doing work, managing tasks, running projects, integrating into tools, and working across sessions.

So here’s the real question:

What do you have that AI can’t copy?

Because those are the only things the next model release can’t replicate.

Everything else? You’re renting it.

And one last thought:

If engineers are moving from writing code to managing AI systems, is that the future you’re excited about, or the one you’re worried about?

— Aashish

Everyone talks about AI, but no one has the time to learn it. So, we found the easiest way to learn AI in as little time as possible: The Rundown AI.

It's a free AI newsletter that keeps you up-to-date on the latest AI news, and teaches you how to apply it in just 5 minutes a day.

Plus, complete the quiz after signing up and they’ll recommend the best AI tools, guides, and courses — tailored to your needs.

Sign up to start learning.