|
Hello Full Stackers!
Welcome back to Full Stack Friday, fresh off the griddle for product people. 🥞
| 👋 We’ve got 2,492 new people since the last send. Welcome to the stack!! |
And speaking of the stack: this week we crossed 50,000 subscribers. Fifty. Thousand. If we used the kind of pancakes used to set the world record for tallest stack ever, 50K pancakes would be taller than the Space Needle.
When I started this thing the goal was to be useful to a few PMs who wanted to build. I don't have a clever way to say this, so I'll just say it: thank you. A lot of great stuff is on the way.
Enough of that, let's get into it. Today:
| • | I launched a new (free!) course: Codex for PMs | | • | Claude Fable 5, the first truly next-gen model in a long time | | • | Codex goes full work agent | | • | Other reads + the meme |
Here we go:
| 🚀 New free course: Codex for PMs |
Yesterday I launched Codex for PMs, a free hands-on course where, like all my courses, you learn IN the actual tool, building real things. Codex is a great tool now, and it runs on the ChatGPT subscription you might already have, with higher and more transparent usage limits than what you get elsewhere (looking at you Anthropic 👀). I’m still a fan of Claude Code, but these tools are converging fast, everyone’s subscription situation is different, and I want my coverage to be good enough that you can navigate any of them. So I’ve made a decision: all my future content will work across all the major agentic tools — Claude Code, Codex, Cursor, and Antigravity.
Some of what's coming:
| • | LLM Memory and Wikis (how to make your agent actually remember things) | | • | Advanced Personal OS (run your whole life out of a vault) | | • | And some fun ones: travel planning and financial planning (really!) |
Related to that...
| 🚀 Upcoming Mastery Update |
I'm hard at work porting my Claude Code Mastery program to work on Codex and Cursor, which will coincide with a big launch I'm making across The Full Stack PM, and will include a price increase from $200 to $300.
But if you're already using Claude Code, the course is available right now! We've got 750+ product people from all the top companies (Google, Meta, Spotify, Shopify, Airbnb, NVIDIA, you name it) and lots of startups massively upgrading their skills right now.
Claude Code Mastery The deep end: 8 full modules of Claude Code, taught IN Claude Code. • 750+ students from Google, Meta, Spotify, Shopify, Airbnb, NVIDIA, and more
• Active Slack community + weekly live office hours with me
• $200 today → $300 at the cross-platform launch (~2 weeks)
• 14-day money-back guarantee, no questions asked |
| 🤖 Fable 5: Straight up magnificent |
It’s been a long time since a new model felt truly next-gen. We’ve had a year of solid increments, plus the Opus 4.7 debacle (and 4.8 to an extent). Then Anthropic dropped Claude Fable 5 on Monday.
Straight up: it's magnificent.
The GPT-3.5 to GPT-4 jump feeling again. I've been throwing huge tasks I'd normally feel like I needed to be deeply in the loop for, and it's executing them flawlessly:
| • | It built a 14-page SEO guide section for one of my course sites in a single session: research, writing, and an adversarial fact-check pass that caught real errors before publish (things like wrong stat and outdated claims). This stuff just works now. | | • | It found and fixed a sitemap bug I'd been shipping since January. Five months of deploys carrying a stale sitemap, and it spotted it on its own. | | • | I described an automation I wanted (recurring tasks that turn into delegated agent work) and it designed, built, and verified the whole thing end to end the same day. |
And it's crushing the overall vibe check. People love this model.
That said, caveats. It's an expensive model: 2x Opus pricing, and you're going to want to use it a lot. This is the first time I've had to even think about usage limits on the 20x Max plan.
And it IS Anthropic, so there’s the usual extra bullshit. The new safety classifiers can randomly block your access when they decide you’re up to something sketchy, like anything related to biology or cybersecurity, and silently switch you back to Opus 4.8. I had this happen multiple times for routine things.
And, of course, stuff like this:
It’s only included on subscription plans until June 22. After that it goes on the meter.
Still, OpenAI and our friends in China will help push the cost down. The next generation of models is here and the next capability jump has arrived. For me, it feels like it's time to think bigger.
Want to get the most out of it? I've found these useful:
| 🔮 Codex goes full work agent |
While Anthropic shipped a better engineer, OpenAI spent the window repositioning Codex for everything else.
The June update:
| • | Six role plugins: data analytics, creative production, sales, equity investing, investment banking, and yes, product design. Each bundles dozens of apps and skills for that job. | | • | Sites: Codex builds, deploys, and hosts dashboards, internal tools, and web apps on OpenAI's own infrastructure. Idea to live URL with zero repos and zero deploy config. | | • | Annotations: mark up documents, spreadsheets, and slides and have Codex refine them in place. |
5M+ weekly active users | | 6x growth since February | | 6 role plugins |
As agents continue to improve at knowledge work, the next battleground is the surface where that work happens. OpenAI’s play is to package by role (your job title gets a plugin) and own distribution (your output lives on their infra). Anthropic is running the same play in other domains: their Claude for Legal stack is up to 20+ connectors, 12 practice-area plugins, and 90+ prebuilt agents.
The takeaway: The durable skill is knowing how to work with agents (delegation, review, decomposition) across whichever surface wins your company. |
And yes! This fits very nicely into the narrative of why I built Codex for PMs
mattpocock/skills: Matt Pocock published his personal agent skills and the repo went vertical: 22k GitHub stars in 24 hours, 80k+ now. The fascinating part is what went viral: skills that add friction. grill-me interrogates your plan before any code gets written, and tdd enforces test-first. The best agent upgrades right now are process constraints that get you more in the loop.
When AI builds itself: Anthropic’s essay on recursive self-improvement, with the most credible number yet on AI-native engineering: Claude now writes 80%+ of the code merged into Anthropic’s own systems. Read it with the obvious caveat that it dropped days before they launched a model. Still, if the number is real, the “AI writes most of our code” era is kicking into high gear.
Hermes: it’s worth a look purely for the memory architecture: a local SQLite search index plus plain markdown files that stay loaded in context. No vector database or cloud. If you’re designing memory for anything agent-shaped, this is the pattern the community is converging on. (Sorry OpenClaw.)
It's all Fable memes this week
Despite all the changes, your connection to all these advancements is going to be through an AI agent.
I now have intro courses for all the major agents. Pick whichever one matches your subscription and do Module 1 this week if you haven't already. It takes about 3 hours, and you'll come out the other side massively unlocked.
Here's a real message I got literally as I wrote this piece:
If you already have the hang of these and you like my style, check out my Mastery course! 8 entire Claude Code IN Claude Code modules, 750+ students, an active Slack, and still on discount until the cross-platform launch lands in two weeks.
Claude Code Mastery is $200 until the cross-platform launch, then $300, with a 14-day money-back guarantee. Join here.
Keep building,
Carl 🥞
|
