Hello! Welcome back to Full Stack Friday, serving up the hottest AI developments fresh off the griddle for product people. | 👋 We've got 8,635 new subscribers since the last send. Welcome to the stack!! 🥞 |
I've been a little quiet here, but I have been busy. After my post about Claude Code for PMs: Mastery went viral, all my energy has gone into building the course, beta testing, launching with the Founding Class, iterating, and building a platform to scale it, plus working through almost 1,000 applications. But we're on the other side. I'm back and have so much in store. More on the course below. Six weeks is an eternity in this space, so there's a lot to catch you up on. At the same time, everyone is basically building the exact same thing! So it's also not as much as you might imagine. Let's do this. |
| • | CC4PMs Mastery is generally available, final week of $200 pricing | | • | The Team OS Workshop is back with Hannah Stulberg, new and improved | | • | The meta-story: every major lab shipped the same thing | | • | Sizzle reel for Anthropic, Google, OpenAI, and Cursor, every major release condensed |
|
🤖 CC4PMs Mastery is Officially Available I'm excited to announce my Claude Code Mastery course is officially generally available! After rolling it out for the last 6 weeks and making improvements with the Founding Class, it's ready for primetime. Master Claude Code for product work. Build research systems, product docs, data workflows, and your own PM operating system, one lesson at a time. The course builds on my "Claude Code IN Claude Code" interactive method of teaching. You learn by doing, fully hands-on inside the tool itself, not videos or slides. Curriculum: 35 lessons across two tracks 🛠 Power User Track (19 lessons), the foundation | • | Core: persistent memory, context management, sub-agent orchestration, custom skills, MCP integrations | | • | Skills: turn any workflow into a reusable slash command, master the 5 skill types, build skills that learn from feedback | | • | Personal OS: study real architectures, build your PM workspace, connect real tools and data | | • | Org OS: make skills team-ready, shared toolkit repos on GitHub, governance practices |
|
🎯 PM Track (16 lessons), practical application | • | Research: executable research plans, interview insight extraction, competitive analysis | | • | Builder: navigate codebases, make targeted product improvements, spin up prototypes | | • | Data: explore datasets, diagnose metrics, design A/B experiments | | • | Docs: AI-powered documents, simulate stakeholder reviews, live-data presentations |
|
And on top of that curriculum I made it an even better offer: | • | All future courses as long as you're subscribed. That means every Claude Code update, plus AI Agents & Orchestration, Advanced Vibe Engineering, and AI Fundamentals (all coming in the next few months). | | • | Private Slack community with 500+ PMs from Google, Meta, Microsoft, Airbnb, Spotify, Shopify, Zalando, Yelp, and pretty much every other top company. I'm in there every day. Members keep telling me it's one of the most active communities they've been in. | | • | Weekly live office hours with me across two time zones. Bring what you're working on or questions you have and get my take, plus input from other AI-forward product people (many of whom know more than me about certain subjects!). | | • | 55-page reference site + office hours recordings + course downloads. | | • | 14-day money-back guarantee. Cancel anytime. |
Early applicant pricing $200/year Use code EARLYAPPLICANT | Goes to $300/year next week. | Claim it → |
|
Two notes: | 1. | Lots of people are expensing this with company budgets. You'll get a receipt after purchase, and if you need anything else to help with your company's process, just let me know and I'll help you! | | 2. | Let me know if you'd like to enroll your whole team for a small team discount. |
🏗️ Team OS Workshop (with Hannah Stulberg) We ran this workshop two weeks ago and it was a huge hit. This is the new and improved version, with everything we learned baked in. AI productivity gains have a natural ceiling. At some point you can't move faster than your team. You move at 10x speed but every handoff is still a 30-minute "let me catch you up" call. Your teammate can't find the call notes you automated. The fix is a Team OS: a shared place where your team's knowledge, workflows, and automations live so Claude can read from it whenever anyone works. Four buckets: | • | Context is the data Claude reads (research, metrics, playbooks, templates) | | • | Actions are slash commands, skills, agents, workflows | | • | Behavior is how Claude should act (naming conventions, review checklists, hooks) | | • | Connections are Slack, GitHub, your data warehouse via MCP |
My co-host Hannah Stulberg is building this for real at DoorDash right now. The workshop is the playbook she's running on her own team. Team OS Workshop June 3, 2026 5:30–8:30pm EDT · $200 · Recording included With Hannah Stulberg (PM @ DoorDash) + me | Hosted on Maven | Register → |
|
|
🌐 Every Lab Shipped The Same Thing |
The single most interesting thing about the last 6 weeks: Anthropic and Google literally both launched products called "Managed Agents." Same exact name, same week! And all four major companies (add OpenAI and Cursor) converged on the same architecture: a model + an orchestrator + persistent cloud environments + mobile/remote control + native MCP. 2025 was the year of the chat agent. 2026 is the year of the place where agents live. |
The nice thing is you don't have to learn every single one of these platforms. The most important thing is to think about the architecture and structure of how these agents work, not the individual platforms. The trend so far is that, given enough time, they all converge on the best patterns. I have a lot of content coming out in this space soon. Now let's go through what each one shipped. |
🤖 Anthropic | • | Claude Opus 4.7. Same price as 4.6, CursorBench 58→70%, new xhigh effort tier, new tokenizer, vision up to 3.75 MP. I really hate 4.7. They ChatGPT'd my boy and I have seen zero improvements. Too much bechmarkmaxxing I think. You can switch back to 4.6 (with 1M context) using this command: /model claude-opus-4-6[1m] | | • | Managed Agents: Dreaming, Outcomes, Multiagent Orchestration, Webhooks. Dreaming is a scheduled process where agents review past sessions and curate memory between runs. Outcomes adds a grader agent that scores work against a rubric and loops until it passes (up to +10 points on internal benchmarks). Multiagent lets a lead agent delegate to specialist subagents on a shared filesystem. | | • | Routines in Claude Code: scheduled and recurring tasks. Think GitHub Actions, but agent-native. | | • | Bonus:/ultrareview slash command for deep code review, Auto Mode extended to Max users for longer unsupervised runs, Claude Design in Anthropic Labs (prototypes, slides, and one-pagers from prompts, pitched explicitly at PMs without design backgrounds). |
✨ Google | • | Google I/O 2026. Pichai framed the whole thing around the "agentic Gemini era." Numbers worth keeping: 900M MAU on the Gemini app (up from 400M a year ago), 3.2 quadrillion tokens/month, ~$180–190B capex this year. | | • | Gemini 3.5 Flash. A Flash model that beats the previous Pro tier on most benchmarks (Terminal-Bench, GDPval-AA, MCP Atlas, CharXiv). 4x faster than other frontier models at less than half the cost. Now the default in the Gemini app and AI Mode globally. 3.5 Pro coming next month. | | • | Antigravity 2.0. Standalone desktop app + CLI + SDK that orchestrates multiple agents in parallel. Powered by an Antigravity-flavored 3.5 Flash that's 12x faster than other frontier models. A direct shot at Claude Code and Cursor. | | • | Gemini Spark. A personal 24/7 agent on dedicated Google Cloud VMs. Connects via MCP. Operates inside Chrome as an agentic browser later this summer. | | • | Gemini Omni Flash. Native multimodal generation (video from text, image, video, or audio inputs). Image and text outputs coming. | | • | Managed Agents in the Gemini API + native Android vibe coding in AI Studio. Yes, same name as Anthropic. Same week, even. |
Definitely giving these vibes: 🔮 OpenAI | • | GPT-5.5 + Pro tier. Brockman framed it as the "super app" foundation. Step-change on agentic coding, computer use, and knowledge work. Uses significantly fewer tokens than 5.4 on Codex tasks at matching latency. | | • | Workspace Agents. Codex-powered enterprise agents that plug into Slack, Salesforce, Notion, Drive, M365, Atlassian Rovo. This officially ends Custom GPTs as the enterprise primitive. Agents are the building block now. | | • | Codex became a "full agent workspace" with a /goal lifecycle, browser verification, an approval-review system, a plugin architecture, and mobile remote control. Scan a QR code from your desktop and approve or redirect agents from your phone. | | • | Bonus:ChatGPT Images 2.0 (multi-image reasoning), Voice API trio (Realtime-2, live translation across 70+ languages, streaming Whisper), and Sora consumer app shutdown (API stays until September). |
🎯 Cursor | • | Cursor 3 → 3.1 → 3.2 in 3 weeks. Anysphere rebuilt the IDE around an "Agents Window." The file tree is no longer the default view. 3.1 added tiled multi-agent layouts and voice. 3.2 added /multitask async subagents, expanded worktrees, and multi-root workspaces. 35% of Cursor's own merged PRs are now agent-generated. | | • | Composer 2.5. Cursor's in-house frontier coding model, trained on 25x more synthetic RL tasks. Built on top of Moonshot's open-source Kimi K2.5 checkpoint. Same pricing as Composer 2 ($0.50/M in, $2.50/M out). | | • | Co-training a 10x larger model from scratch with SpaceXAI on Colossus 2's million-H100-equivalents. This is the under-covered story of the window. Cursor isn't a wrapper anymore. They're betting they need to own the model layer to win the orchestration layer. | | • | Cloud agent development environments. Configurable cloud envs with cloned repos, dependencies, internal toolchains, multi-repo support, and programmatic triggers for autonomous runs. |
|
At the end of the day, the way you're going to work with any of this stuff is by using a coding agent. If you haven't done one of my free courses, start with whichever fits your tool of choice: If you have, check out Mastery. I'd love to have you in the program. Final week of $200 pricing before it goes to $300. Use code EARLYAPPLICANT. Keep building, Carl 🥞 |
|
