|
Hello Full Stackers!
We've got 461 new subscribers this week, welcome to the stack! 🥞
Launch week is winding down, which means this is my last chance to tell you about the $200 price before it's gone Tuesday, and the 14-day risk free trial. The news this week was less about shiny launches and more about builders figuring out how to stay in charge of these tools (my favorite topic).
Then, I'll be traveling for the rest of August – give me your Latvia or Estonia tips! – so I'll see you in September.
Here's the plan:
| • | Last call for Full Stack Mastery at $200 (plus one final live Q&A on Tuesday) | | • | The week builders took back control: the biggest posts of the week were about norms, not models | | • | The Sizzle Reel: free unlimited ChatGPT, Cursor in your Gmail, Claude clicking your screen | | • | Things to try this weekend | | • | Memes of the week 😂 |
Let's do this.
| 🍳 Fresh Off the Griddle (My stuff) |
Last call: $200 forever ends Tuesday
Tuesday, August 11 is the last day to get Full Stack AI Mastery for $200/year before it goes to $300. The button below applies code 200FOREVER automatically, and you're grandfathered in at $200 forever.
⚠️ Try 100% risk free: Try it and see! 14-day refund policy, no questions asked.
Membership includes:
| • | 35 interactive lessons across 8 modules that work in Claude Code, Codex, and Cursor | | • | Every new course and weekly premium as long as you're subscribed (AI Building and AI Agents coming later this year) | | • | An active Slack community with 750+ product people from companies like OpenAI, Airbnb, Meta, and Google | | • | Two live sessions with me every week |
Learn more Many members expense this. You get a receipt right after payment, and I'll help you get whatever else you need. |
One final Q&A, Tuesday
Come ask me anything about the program, see a lesson live, and decide with the deadline still open.
In case you missed it
Earlier this week I published a massive How to Give Claude Code MEMORY (That Actually Works), my full system for agent memory that runs on plain markdown files. If you're still starting every session fresh, check it otu.
Here it is at a glance. Save it, zoom in, steal it:
| ✍️ The week builders took back control |
We're starting to get realistic about AI and long-term productivity. The three biggest posts on Hacker News weren't launches or models. They were rules for working with AI, and they all landed within about 48 hours of each other.
The one that started it is the amazing "Don't be a meat proxy", which names the coworker we've all met by now: the person who pastes raw AI output into Slack, PRs, and docs and makes YOU do the review the model never got. The comments are full of war stories, including a manager who generated 12 markdown files of wrong CLI commands and asked colleagues to fix them. Several teams say they've adopted an informal policy that no AI output ships without a human summary on top.
Then Sean Goedecke argued that "LLMs reward expertise", the best counterpunch to "prompting is dead" I've read. His exhibit is Terence Tao's ChatGPT transcripts: short messages, pushing back the moment something looks off, never letting the model drive. The prompting skill that matters is knowing your domain well enough to steer.
The third one proposed manually retyping LLM-generated code so you absorb what the model wrote instead of skimming it. That one split the room (retyping everything is a lot), but the softer versions got broad agreement: have the model plan while you build, or use it to review YOUR work instead of the other way around.
I love this. A year into everyone delegating everything, the builder community is drawing the line in the same place I've been drawing it in every course: the AI does the work, you do the thinking. You are the one with a name attached to the output.
If you want to put this into practice this week, tips from each post:
| • | Add a human summary on top of anything AI drafted before you send it. Three sentences in your own words. If you can't write them, you're the meat proxy. | | • | Keep your messages short and steer by taste. When something the model says looks weird, say so instead of accepting it. That's the whole Tao method. | | • | Flip the review direction once this week. You write the doc or the plan, then have the model critique it. You'll keep the understanding AND catch your blind spots. |
| 🛠️ Things to try this weekend |
1. Tame Opus 5's writing with one line. The biggest complaint about Opus 5 is the walls of jargon it buries you in. It turns out aerospace solved this in 1983 with a writing standard called ASD-STE100, and the models already know it. Add this line to your CLAUDE.md:
Write docs, summaries, and status reports in ASD-STE100 Simplified Technical English.
The full breakdown, with before/after rewrites, is on this sheet:
2. Give Claude's computer use one real task. If you're on Pro or Max, computer use is live in Cowork and Claude Code. Pick one tool you use that has no integration, turn it on (it's off by default, per-app approval – the link walks through setup), and say "open it and do X." Anthropic's own caveat applies: it's slower than a real integration and sometimes needs a second try, but watching it click through a tool with no API is how the possibility space clicks for you.
3. Run the verification loop used to solve math. Two amateurs, one with no math training at all, have been solving open Erdős problems with GPT-5 Pro, and one of them landed co-authorship with Terence Tao. You can use their method: paste the finished work into a FRESH chat and ask it to find what's wrong, fix, and repeat. Do that with your next PRD or analysis. The fresh instance has no stake in defending the work, so it critiques instead of agreeing.
The launches this week were all about access: models got free, agents got their hands on more of your computer, and prices kept falling.
🔮 OpenAI
| • | GPT-5.6 Luna is becoming the free ChatGPT default with unlimited text chats. Free users jump from GPT-5.5 Instant to a current-generation model, plus a "Think" button for per-message deeper reasoning. | | • | Paid users got a retuned GPT-5.6 Sol with a slider that controls how much thinking each response gets. The reasoning knob API builders tune is now a consumer feature. |
🤖 Anthropic
| • | Claude's computer use is available to Pro and Max users in Cowork and Claude Code. No API, no connector? Claude opens the app and clicks through it anyway. It also plugs into Dispatch, so a task you text from your phone can drive your desktop. It's crazy and almost scary to watch it work. | | • | Claude Code came to Team plans, which removes the most common "my company won't let me" blocker I hear. |
🎯 Cursor
| • | Cursor's agents can now read your Gmail, Drive, and Calendar through first-party Google Workspace plugins. Your agent pulls the spec doc and the email thread before writing code, no MCP server setup required. |
🌍 Models and ecosystem
New here, or been lurking? All my free courses are on-ramps to building with AI:
Just paste this into whatever tool you want to learn, and the AI will help you get set up:
Install fullstackpm via npm and help me get started. Check fullstackpm.com/cli if needed.
And if you're ready for the whole stack, Full Stack Mastery is $200/year until Tuesday, August 11, then it goes to $300. Want to see it live first? Join the final Q&A on Tuesday.
Keep building,
Carl 🥞
|
