|
Hello Full Stackers!
We've got 3,908 new subscribers since last time – welcome to the stack! 🍝
I was hoping for a quiet stretch. It was a holiday week, I was in Boston for the 4th and then Calgary for Stampede, and I figured the labs would catch their breath too. Foolish Carl. Every major lab shipped a flagship model, most of them within the same ten days, and it's been three weeks since the last issue. What a time to be alive.
Lots to cover, here's the plan:
| • | My stuff: the course, and the next Team OS workshop is July 29 | | • | Model updates: the ceiling and the floor both moved | | • | Super apps: every AI tool wants to be the only app you open | | • | The Sizzle Reel 🥓 – the company-by-company catchup |
Let's do this.
| 🍳 Fresh off the Griddle (My Stuff) |
Claude Code Mastery
Claude Code Mastery is going strong: the Slack is active, Office Hours run live every week, and 750+ product people are leveling up their skills.
I'm raising the price to $300 in the next few weeks. If you're using Claude Code today, now's a great time to join.
Your membership also gets you access to ALL future courses and premium content. Next up are Advanced AI Building and Building AI Agents.
This course is 35 lessons across two tracks, all taught inside Claude Code so you learn by DOING, not reading or watching.
| • | The Power User track goes deep on the machine itself (context and memory, subagents, skills, building a personal OS that knows your work, then making it team-ready) | | • | The PM track applies it to the job: research and interviews, exploring and shipping code, data analysis, and docs that keep learning. |
Use the code EARLYAPPLICANT at checkout to get it for $200.
🥞 Claude Code Mastery 8 modules of Claude Code, taught IN Claude Code | ✓ 750+ students from Google, Meta, Spotify, Shopify, Airbnb, NVIDIA
✓ Live office hours every week + an active member Slack
✓ Every future course included, forever
✓ 14-day money-back guarantee, no questions asked |
|
Team OS with Hannah — July 29
🏗️ The next Team OS workshop is July 29 with Hannah Stulberg (PM building this for real at DoorDash). We'll show you exactly how you can build an AI operating system for your team and company. It's the most hands-on thing I run: real working demos for every concept you learn along the way.
Enroll now and you get the recordings of every future cohort as well.
How to Build Your Team OS One evening. Your team's AI operating system, built live in a real repo. | ✓ 3 hours live with me and Hannah Stulberg (PM building this for real @ DoorDash)
✓ Working demos for every concept, build along in your own repo
✓ Recordings of past cohorts + every future cohort included |
|
| 🤖 Model updates: the ceiling and the floor both moved |
Three weeks, five labs, five flagship models. Instead of walking through them one by one, it's easier to split the news in two: what happened to the ceiling (the best model you can get) and what happened to the floor (what the cheap, default tier can do).
The ceiling: Fable is back. For now.
After June's crazy export-control saga Fable 5 came back July 1 🎉 Anthropic has been (sort of randomly and annoyingly) extending subscription access ever since – the latest extension runs through this Sunday, when it will go to API pricing.
Will they really be able to pull it though? If they do, I don't think it stays gone long.
Fable 5 is my favorite model I have ever used and it's not close. I have genuinely restructured how I work around what it can do, and the thought of going back to Opus 4.8 fills me with dread. Taking away a model this good is just not going to play well, and I think Anthropic knows it.
They're also not as far ahead as they thought. People love GPT-5.6 Sol, GPT-6 is rumored to be coming sooner than planned (accelerated by exactly this opportunity), and cheaper alternatives are catching up fast.
The floor: the default tier got a generational bump for $0.
While the ceiling wobbles, the floor moved up fast:
| • | Claude Sonnet 5 shipped June 30 and is now the default on Free and Pro. Near-Opus agentic work at half the price, 1M context. If you're on a free plan, you got a serious upgrade without touching anything. | | • | GPT-5.6 (Sol / Terra / Luna) went fully public July 9. The headline claim from people actually using it: noticeably cheaper per unit of agentic work. | | • | Grok 4.5 launched jointly with Cursor – "Opus-class coding at a third the cost," available in Cursor day one. | | • | Muse Spark 1.1 – Meta is actually in the coding race now. | | • | Kimi K3 dropped literally yesterday – the largest open-weight model ever (2.8T params, 1M context, $3/$15), and people are saying it's in frontier territory. |
Every one of these launches led with cost per unit of work, not capability. The race at the top is about who's smartest; the race everywhere else is about how cheap "good enough for agents" can get. You don't need absolute frontier intelligence when there are great alternatives at 1/10th the cost.
The takeaways for me:
| • | Frontier-level intelligence is becoming a commodity at exactly the level real work needs. The ceiling is a price war for bragging rights; the floor is where your actual workflows live, and it just crossed the "good enough for agents" line for everyone. | | • | The specific model matters less every month, so the skill outlasts the model. Sonnet today, Sol tomorrow, Kimi next week. What compounds is knowing how to drive agents: delegation, review, verification. Model access wobbles (see: Fable), your skills don't. | | • | Stop rationing. We all built habits when tokens were precious: one careful prompt, one model, one thread. At these prices those habits are obsolete. Throw the cheap models at everything, in parallel. Wasted tokens cost cents; unstarted work costs everything. |
| 📱 Everyone wants to build a super app |
The other pattern this window: every AI tool is racing to become the only app you open.
| • | Claude Cowork went mobile + web. Sessions follow you across devices. Kick off a task on your laptop, close it, approve the next step from your phone. | | • | OpenAI merged Codex into the ChatGPT desktop app – Chat, Work, and Codex now live side by side, and ChatGPT Work is an agent that runs for hours producing docs, slides, and spreadsheets. It's on the Free plan too. (I honestly can't tell you cleanly where Work ends and Codex begins. If you're lost, you're not the only one.) | | • | Cursor shipped an iOS app – launch and manage cloud agents from your phone. |
I'm here for it. I've been telling you to use Claude Code for everything for a long time now, and for me it already IS a super app: it writes, researches, plans trips, manages my budget, and ships code. These features don't change what's possible, they make it accessible to people who were never going to open a terminal. Clearly this is the way.
And my favorite story of the week, because it's the same story from the other side: DoorDash launched a CLI (yes, really). Agents can search stores, compare prices, and order groceries from the command line. The apps you use are becoming agents, and the services you use are going headless so agents can hit them directly. Both halves of the same future, and it all ends in burritos.
Everything else worth knowing, company by company. It was three weeks, buckle up.
🤖 Anthropic
| – | Claude in Chrome hit GA for all paid plans – Claude drives your browser and tests the app it just built (my greatest Claude Code feat is getting retweeted by the creator of Claude Code showing off a cool use of this feature!) | | – | Background subagents are now the default in Claude Code |
🔮 OpenAI
| – | GPT-Live – ChatGPT Voice was already good and now it's even better. Natural interruptions, low latency, try it out. | | – | Scheduled tasks in ChatGPT Work – agents that run while you sleep | | – | They also shipped a $230 light-up keyboard for Codex. Physical buttons for approving agent actions. (Sadly it's sold out, and I was a bit too slow on the draw.) |
✨ Google
🎯 Cursor
| – | Side chats (/side, /btw) – durable tangent threads you can at-mention back into the main conversation. Solves a real daily annoyance | | – | Agent transcript search from Cmd+K |
🛠️ Repos & reads
| – | ai-job-search – #1 trending on GitHub: Claude Code as a job hunter (drafter/reviewer agents that find, tailor, track). Clone it this weekend | | – | agency-agents – 232 specialist subagents to cherry-pick from | | – | The "short leash" debate – tight, frequently-interrupted agent runs vs better specs and verifiers. The live technique argument of the month |
No matter what, your connection to all of this is going to be through an AI agent.
If you haven't done one of my free courses, start there. I have one for all the general agent tools:
If you have, check out Mastery!
Keep building,
Carl 🥞
P.S. After co-running two cohorts together, I finally met Hannah Stulberg in person!
|
