Week in Product #494 🚀
Hermes course, GPT-5.6 preview, Claude Tag, New Figma features, New Meta glasses, Experimentation in the AI era, Building a company OS, How to "Platform-PM" & more
Hi folks 👋
Welcome to a new Week in Product!
🎰 The week in figures
$5.4B: World Cup betting hits $5.4B across Kalshi, Polymarket, and Robinhood prediction markets
$1.5B: Baseten is raising $1.5B at $13B valuation to expand low-cost AI model services competing with OpenAI, Anthropic on affordability
$650M: Groq raised $650M to scale AI inference cloud to 200 MW by 2027 across 13 global data centers serving 5M developers
$500M: RAISE US launched with $500M+ raised and a goal of $1B, backed by OpenAI, Anthropic, Amazon, Microsoft, and Bank of America, to fund retraining programs for workers displaced by AI, starting in Arkansas, Connecticut, Maryland, and Utah
$320M: General Intuition raises $320M Series B at $2.3B valuation for using gaming to develop embodied AI, robotics, spatial intelligence, pushing world-model research
$320: Airwallex has announced a $320M Series H at $11B valuation to accelerate product development across autonomous finance and agentic commerce, while also expanding its infrastructure and regulatory footprint into new markets
75%: Credit card data shows Anthropic's paying consumer base up 75% since Jan-26, with interest in Claude courses surging 18x in 30 days, gaining ground on OpenAI
5x: OpenAI's Codex grew its active user base more than 5x in the first half of 2026, with non-developer adoption rising 137x for individual users; legal, finance, and recruiting at OpenAI now use Codex as their primary work tool, not ChatGPT
📰 What’s going on
OpenAI holds back GPT-5.6 series and announces a “limited preview” of its new frontier AI models. That includes Sol, the flagship; Terra, a comparable model to GPT-5.5 that eats up only half the tokens; and Luna, an even lower-cost option. For now, only a few trusted partners have access to these models, as requested/instructed by the US government. The models will apparently open up access to a larger population within the coming weeks
OpenAI leadership is having second thoughts about its IPO plans, at least for the remainder of 2026. Sam Altman wants the value to exceed $1T as a public company, and it’s unclear if that’s currently possible amid unpredictable global markets, lingering concerns about future AI revenues, and the downward slide of SpaceX stock following its own blockbuster debut
OpenAI and Broadcom team on AI chip. This marks the first-ever custom AI chip developed by the ChatGPT maker. For now, Jalapeño remains in the testing phase, but Broadcom reports that the chip provides equivalent inference performance while draining about 50% less energy, compared to a standard AI GPU
Anthropic has launched Claude Tag, a new service that embeds an "always-on" AI assistant directly into Slack channels, where it acts as a persistent teammate that learns from ongoing conversations and can complete multi-step tasks autonomously. The feature is now in beta for Claude Enterprise and Claude Team customers. Andrej Karpathy referred to such services as a “self-contained, persistent asynchronous entity with org-wide tools and context, working alongside teams of humans”, which he calls the next iteration of LLM UX
Anthropic accuses Alibaba of the largest-ever Claude distillation attack, alleging Alibaba ran 28.8M fraudulent exchanges through 25K accounts between April-June 2026 to illicitly extract Claude's capabilities and accelerate China's AI development
Google delays Gemini 3.5 Pro to July for token optimization. The company pushed its next frontier model back from June to gather more real-world feedback and fix token consumption issues identified in Flash 3.5
Meta releases new smart glasses without Ray Ban frames, at $299 with camera, AI assistant, 8-hour battery, & 14 live-translation languages
Meta paused its Model Capability Initiative (an internal program that records employee keystrokes, mouse clicks, and screen activity to train AI) after a permissions error made all that sensitive data readable by everyone at the company
Instagram enhances its TV app with longer-form episodic and live formats now on Samsung TVs
Apple announced sweeping price increases across Macs, iPads, and HomePods blamed on chip shortages. The company said it has "never seen a component price increase this much, this quickly." Updated prices:
MacBook Neo: $699 (up from $599)
13-inch MacBook Air: $1,299 (up from $1,099)
13-inch MacBook Pro 1TB: $1,999 (up from $1,699)
Base iPad: $449 (up from $349)
iPad Air (11-inch): $749 (up from $599)
Vision Pro: $3,699 (up from $3,499)
Figma announces Motion, a native animation timeline built directly into Figma. Designers can now animate objects on the canvas with keyframes, easing controls, and sequencing without leaving the file or handing off to a separate tool. Here’s an example of one animation that took 4.5 seconds to make
Figma also presented Generative Plugins, which let you ask the Figma agent to build a plugin in plain language without knowing how to code it. The plugin lives in the file and travels with it, so if you hand a file off to another team or a client, the tools you built with go with it
And finally, Figma announced Code Layers. Code now lives directly on the canvas, so designers can import a GitHub repo or use a local folder, and when making changes that impact the codebase, Figma’s agent analyzes what’s been changed and pushes updates back to the main repo
China claims No. 2 global AI lab ranking: New Zhipu AI's GLM-5.2 model trails only Anthropic's Fable 5 on front-end coding benchmarks; founder Tang Jie predicts China will match Fable 5 before year-end, challenging Musk's Q1 2027 projection
Alibaba's new AI video model HappyHorse 1.1 surges to #2 spot in benchmarks, surpassing OpenAI's Sora and ByteDance's Seedance
Notion announced external agent support for Claude and Cursor
Notion will shutter its Mail inbox later this year, arguing that more of its customers are using agents to handle email Consider this (relatively minor) product decision as one more piece of evidence that future software will be built for agents, not humans
SoFi has acquired Composer, a startup that helps investors build and automate trading strategies using AI
📚 Good reads
The AI era requires a different kind of experimentation, by Elena Verna. In an AI-first world, tweaking button colors and tiny UI details is a distraction; the real gains come from bold, system-level experiments, especially around monetization. It’s time to skip minor tests, run bigger, longer-lived bets on pricing and packaging, and to treat AI “best practices” as the default, so scarce eng time goes into experiments that move engagement and revenue over the long run
Trust before it’s earned. Leah Tharin explores why leaders should give trust early, and argues that real empowerment means granting ownership up front, then demanding no surprises through clear, frequent updates. She breaks down how “managing up” with actionable transparency protects trust, and how servant leaders should never punish failure but must still act fast on weak performance. Strong performers are the ones who reduce chaos, create clarity, and quietly lift everyone around them
Why platform product management is the most demanding PM role. Platform PMs juggle two very different “customers” (internal teams and end users), a swarm of stakeholders, and impact that only shows up months later. Because the work is mostly invisible, they also need exceptional storytelling and influencing skills to secure adoption. Treating platforms as products (with JTBD, clear leading indicators, internal GTM, and sane versioning policies) turns a hard role into a high‑leverage one
🎧 Good listen/watch
[Podcast] How Laurel built an AI-native “company OS” for everyone. Laurel’s CPO, Jiaona Zhang, shows how a simple Claude Code “company OS” lets even CSMs ship to production. It also reframes the PM role in AI-native teams (captain model, two-track reviews) and gives a clear framework for proving you’re an AI-native PM in interviews
🧑💻 Worth learning
45-min Hermes course: set up your AI chief of staff. Peter Yang introduces a beginner‑friendly course to get Hermes running 24/7 in under an afternoon
Claude Code power skills for PMs. This guide shows how to set up agentic workflows in Claude Code. It covers power habits like tight context management, planning before execution, automatic checks, picking the right model (Haiku/Sonnet/Opus) for each task, and wiring hooks, agents, and /goal loops. A brief operating manual for turning Claude into a reusable execution engine
Anthropic has published a detailed guide on how to manage Claude Tag
🔧 Products to try
AgenticCalling lets you send your Claude or ChatGPT agent out to make real phone calls on your behalf, so you can tell it “call 50 hotels in Miami and find me the best rate” and it handles the whole conversation autonomously; 3 free minutes to start
LocalClicky is a fully offline Mac voice assistant that controls your computer by voice (open apps, move files, edit videos, click things on screen), with zero data leaving your machine; open source, free
That’s all 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!
