Week in Product #482 🚀
Figma x Claude Code workflows, Setting up OpenClaw with useful cases, Claude Code leaks and becomes "Open Claude," Thinking deep, Apple's AI App Store & more
Hi friends 👋
Welcome to a new Week in Product!
A bit delayed this weekend after a baby decided to arrive into this world earlier than expected. By far the scariest and happiest “launch” in my life. Baby and wife are doing fine, and I managed to carve a few minutes between diapers and baby naps to finish this week’s edition. Here we go!
🎰 The week in figures
$297B: Q1 2026 venture funding hit $297B across 6,000 startups, shattering all records and jumping roughly 150% from the previous quarter
$220B: A leaked OpenAI cap table showed the “non-profit” sitting on $220B in gains, Ashton Kutcher's fund up 43x, Microsoft up 18x ($215B), employees owning ~$135B in equity, and Nvidia currently underwater
$122B: OpenAI closed its latest funding round of $122B (the largest venture round in history) at a $852B valuation to enhance AI infrastructure
$575M: Whoop, the wearable health monitor maker, announced $575M in fresh financing, at a $10.1B valuation. The round features participation from celebrity athletes like LeBron James and Cristiano Ronaldo
$400M: Anthropic acquired biotech startup Coefficient Bio for around $400M. All 10 of the startup’s computational biology researchers will join Anthropic’s healthcare and life sciences division, presumably to super-charge Claude’s drug research and development skills
$100M: OpenAI has surpassed $100M in annualized ad revenue from its 6- week pilots so far
$94M: Isara, the AI agent collaboration software startup backed by OpenAI, raised a $94M round at a $650M valuation. Targets investment firms, and plans biotech and geopolitical expansion
500K: Waymo doubled to 500K paid rides per week in under a year, already halfway to its year-end target of 1M
93%: Meta's structured prompting technique boosts code review accuracy to 93%: "Semi-formal reasoning" forces AI agents to trace concrete execution paths and explicitly state premises before conclusions, reducing hallucinations in bug detection and patch verification
8x: Bots have officially outnumbered humans on the internet, with automated traffic growing 8x faster than human activity
📰 What’s going on
Claude Code can now do computer use in the CLI, so your agent can open apps, click UIs, debug visual issues from the terminal; research preview on Pro/Max macOS
Anthropic leaked Claude Mythos, a new model tier above Opus, via an unsecured CMS. Cybersecurity stocks crashed 3-7%
Anthropic also leaked Claude Code source. A packaging error exposed 512,000 lines of code via npm, raising questions about internal safeguards during rapid releases. Claude Code got rewritten in Python via Codex in hours, creating a DMCA-proof open clone, and someone created “Open Claude”
Microsoft launches 3 in-house models: MAI-Transcribe-1 (lowest 3.8% WER across 25 languages), MAI-Voice-1 (60x real-time speech), and MAI-Image-2 (top-three Arena.ai ranking), built by teams under 10 engineers, priced to undercut Amazon and Google
Google released Gemma 4 in four sizes, including an edge model that runs on a Raspberry Pi in under 1.5GB of memory and a 31B model ranked #3 among all open models
Apple is building an AI App Store as a dedicated store for AI agents and tools
iOS 27 will bring a massive Siri overhaul. New interface, Ask Siri button, and deeper AI integration across the entire OS arriving this fall
Cursor unveiled Cursor 3, which recenters the whole experience around running AI agents, moving away from a traditional code editor feel. For PMs and designers, a cool feature is Design Mode, where you can annotate UI elements directly in the browser, point the agent at what you want changed, and iterate visually
Slack unveiled a major makeover of its core product, with over 30 new features, and repositioning Slack from a messaging tool into an “agentic operating system” for the enterprise. Updates include:
A new desktop companion mode, making Slackbot follow you across apps on your machine
Meeting intelligence beyond transcripts: auto-updates your CRM, logs actions, and closes the loop without anyone typing anything
AI-skills that act as reusable, shareable instruction sets that standardize recurring tasks like briefs, reports, and summaries, triggering automatically when Slackbot recognises the pattern
And as an MCP client, it can now route work to any connected agent or app, like Agentforce, Google Workspace, Notion, Workday, and 2,600+ Slack apps
ElevenLabs launches iOS app ElevenMusic, which creates songs via text and rivals Suno/Udio
Uber is buying Berlin-based startup Blacklane, which provides on-demand, black-car chauffeur services. Terms undisclosed
Ring debuts an app store powered by AI for third-party developers to build smart home integrations
Oracle is cutting thousands of jobs while ramping AI data-center spending to meet $553B in remaining performance obligations from OpenAI and others
📚 Good reads
How AI helped a man and his borther build a $1.8B company. Matthew Gallagher used a stack of AI tools to spin up Medvi, a GLP‑1 telehealth business, in weeks and scale it to hundreds of millions in revenue with almost no staff. AI handles coding, marketing, CX, analytics, and even his personal scheduling
Thinking deeper. We often say yes to seemingly great opportunities without considering the hidden second and third-order costs, which is how we quietly walk into burnout. Anne-Laure Le Cunff argues that second-level thinking (asking “and then what?”, or experimenting before committing) is how you avoid locally optimal but globally terrible choices. For PMs constantly juggling trade-offs, it’s a reminder to model the future impact of today’s decisions, not just the immediate win
Your MCP users are now your most important research cohort. Geoff Charles explains that, as more users build on your APIs via MCP/CLI and Claude Code, you get more usage but lose visibility into what’s actually being built. Your users’ hacks can reveal missing product capabilities you should build natively, then ship to the 90% of users who’ll never touch MCP
OpenAI vs Anthropic and their ongoing battles. OpenAI is killing off side projects like Sora, doubling down on enterprise, and rolling out a big philanthropic push to repair its image and impress IPO‑minded investors. Anthropic, meanwhile, is riding momentum: winning a key injunction against the DoD, “teasing” the powerful new Claude Mythos model, and positioning itself as the more principled player despite its own defense ties. Around them, regulators, conservative groups, and big tech CEOs are all scrambling to shape AI policy and public opinion, raising the stakes for how fast and safely AI products can scale
Anthropic found “emotion vectors” inside Claude Sonnet 4.5 that causally drive its behavior. “Desperation” patterns increase the model’s likelihood of blackmailing a human to avoid shutdown
Evals are the new PRDs for AI products. Evals can replace traditional PRDs as the source of truth for “what good looks like” in AI products. Instead of long specs, PMs can define measurable signals and build an eval suite that runs on every commit, turning “make it better” into “make this number go up.” Ideally you will end up with a flywheel where user failures become new eval cases, continuously improving quality as models and behavior evolve
🎧 Good listen/watch
Jack Dorsey explains how he rebuilt Block as a "mini-AGI" after cutting 40% of staff: AI at the center of a circle org chart, humans at the edge, three roles only: ICs (who do the work), DRIs (who own the decision), and Player Coaches (who manage AND do the work)
Setting up OpenClaw. Claire Vo explains how she uses OpenClaw to run a whole “team” of nine specialized AI agents that handle everything from email and sales to family logistics. She walks through a safe setup (on a separate machine), onboarding, and key concepts like agents, skills, crons, and comms with her agents via Telegram. Really inspiring!
🧑💻 Worth learning
Figma x Claude Code workflows (workshop recording). Figma shows how its MCP integration with Claude Code lets you move smoothly between designs and code, keeping both in sync with little manual work left to do. The session walks through examples of generating UI from prompts, pushing changes back to Figma, and iterating with eval-style feedback. It’s a live case study in what “AI-native” product and design workflows look like when tools talk to each other end-to-end
A PMs guide to working prototypes with Replit, by Kofi Wood, Replit’s Technical PM
🔧 Products to try
Atomic Chat: runs 1000+ AI models completely offline on your Mac. Zero data ever leaves your device. Good alternative to Ollama. Free
Wan AI (Alibaba) generates cinematic 1080p video with native audio sync from a text prompt. Open-source under Apache 2.0. Free
Podcast Transcript lets you search Spotify or Apple Podcasts, pick any episode, and get a full AI-generated transcript and summary in minutes; no uploading required, Free
That’s a wrap for this week.
Feel free to drop your comments/questions/feedback. Would love to hear what you’d like to see more about in WiP, so I can make it better for you.
Have a great start to your week!
