🥞 Opus 4.8 Tries to Save the Day, No Price Increase (yet)!, The AI Vibe Shift

Hello Full Stackers!

Welcome back to Full Stack Friday, fresh off the griddle for product people. 🥞

Quick note: last week I told you the Claude Code for PMs: Mastery price was going up today. I changed my mind! The price is staying put for now, and I'll explain why in a second.

The rest of the week gave us two real stories. Opus 4.8 landed (you may recall I had feelings about 4.7), and the entire industry seems to have woken up with a hangover, suddenly asking out loud whether any of this AI spend was actually worth it.

Today:

Let's get into it.

Last week I told you the price of Claude Code for PMs: Mastery was going up from $200 to $300 today. I changed my mind.

I made a critical decision when I built CC4PMs Mastery. Since these tools update so frequently and the meta around them is always changing, and because I wanted to build real, ongoing community, I decided to make it a membership that includes all my courses instead of a one-off course purchase.

Even in the last few weeks, with Anthropic somewhat dropping the ball, OpenAI crushing it with Codex, and the rise of Hermes as an agent orchestrator, I already know this was the right call.

So I'm holding off on increasing the price of the membership until I have a few more pieces in place, like making the course cross-compatible with Cursor and Codex. Also until I figure out the name: Full Stack PM Pro? Blueberry Pass? Chocolate Chip Tier?? 🥞

Never waste a good price increase, as they say.

That said, CC4PMs is rocking and rolling! We've got 700 product people from all the top companies, an active Slack, and I'm running two weekly office hour sessions every week that always have great discussions. Students are massively upgrading their Claude Code usage for real product work: research systems, PRDs, data workflows, and your own PM operating system.

It's still a great time to join, and you'll be grandfathered into the $200 price forever. Use code EARLYAPPLICANT for the $200 pricing.

The next frontier of AI use in companies is how to build Operating Systems for your entire team. I'm running another cohort on "How to Build Your Team OS" next Wednesday, June 3 with Hannah Stulberg, who is building exactly this for real at DoorDash right now.

We learned a lot from round one, and this version is even more practical, with nearly double the time spent inside real, working demos.

Here's what people said about round 1:

Hope to see you there!

Anthropic dropped Opus 4.8 yesterday, six weeks after 4.7. On paper it beats GPT-5.5 and Gemini 3.1 Pro on most benchmarks, but do we really care about benchmarkmaxxing at this point?

My first impressions: I hated 4.7. 4.8 is clearly more capable, but it still wants to do too much, so deeply wanting everything to be profound and brilliant. And it's a true descendant of 4.7, based on the same new tokenizer, which means our beloved 4.6 personality might be gone forever, and yes, it's token hungry.

As a reminder, you can switch back to Opus 4.6 with this command: /model claude-opus-4-6[1m]. Anecdotally, nearly everyone in my Mastery program switched back from 4.7 to 4.6. We'll see what happens with 4.8.

That said, the way 4.8 works within the harness might be where the real gains come from. This is their new "Dynamic Workflows" architecture:

Basically: On the left is v1 of the "agent team" approach: a bunch of agents talking to each other, negotiating, and hoping it all converges. On the right is a more deterministic workflow: one orchestrator fans out a batch of tasks, and each task runs a tight little loop (implement, then verify, then fix) before reporting back.

This is a good take:

This is why you should be skeptical of every breathless story about someone letting an agent run unsupervised for eight hours, which mostly produces a lot of expensive slop. (More on that in a second.)

Most importantly, it has a fun little easter egg in the UI:

There's been an unmistakable vibe shift around AI recently, really starting to crystallize this week. For two years, just using AI was the whole win and adoption was the metric. But now the bill is coming due and reality is setting in.

This sums it up:

We're also seeing this play out with all the "AI-first restructuring" layoffs at Meta, Cloudflare, and ClickUp. Even my own gripe about 4.7 fits the mood: people are starting to suspect these models are getting more expensive to run, not better.

My take: The era of experimentation is closing, and the era of competence is opening.

Outcomes, and the skill behind them, once again reign supreme. The person who can architect the workflow, manage the context, and actually ship good results will become the valuable seat in the building. This is great news if you're willing to get good.

Seriously, ClickUp is now floating $1M comp bands for the people who can do "100x" work with AI.

What does this look like in practice? A few things:

These are barely even jokes.

In a world where everyone is suddenly asking whether AI is worth it, there's never been a better time to shine.

Start with one of my free courses: Claude Code for PMs, Cursor for PMs, or Claude Code for Everyone. They're still the fastest way in.

Once you've done one of those, come join Mastery for the full curriculum, the Slack, and weekly office hours.

And if you lead a team, the Team OS Workshop with Hannah is June 3.

Keep building,

Carl 🥞

P.S. Last week was a huge week for me personally! 500 paid members for my community has been my "okay-I-am-really-DOING-this" number, and we hit it last week. I'm all in on this for the long haul. So many good things to come. Thank you for being here.

Of course I had to celebrate with pancakes. I swear this is a real image even though for some reason it looks so AI generated. I think I've been Claude Coding a little too much.