Week in Product #495 🚀
Meta glasses subscription, Fable re-launches, GPT 5.6 preview, Cursor and Meta new apps, Summer reads, Grading AI fluency, Using Claude Tags, AI Product OKRs & more
Hi friends 👋
Welcome to a new Week in Product!
🎰 The week in figures
$18B: Bending Spoons, the Italian tech company, achieves $18B Nasdaq IPO with founder sharing "failure playbook" of strategic pivots, lessons from setbacks
$2.5B: Microsoft launches a $2.5B AI deployment unit with 6,000 engineers. It will embed industry experts at Fortune 500 clients for outcome-driven implementations; follows Amazon's $1B FDE commitment two days earlier
$2B: Kling AI, the Kuaishou spinoff, raised $2B at a $15B valuation with $500M ARR; Q1 revenue 650M yuan, marking 300% YoY growth, competing with Seedance on video AI
$800M: Together AI raised $800M at an $8.3B valuation to expand open-model infrastructure
$500M: Higgsfield AI, the San Fran video startup, in talks to raise $300-$500M at a $5B valuation, with $500M monthly revenue run rate up from $200M this year
$320M: Airwallex raised a $320M Series H at $11B valuation, and announced two new products: Airi, a consumer wallet for agentic commerce across fiat and stablecoins, and T:0, an AI-powered stack for business finance
$200M: Quantifind raises $200M round.vThe startup develops AI-powered software to help banks detect and combat financial crimes, particularly money laundering, using a combination of public and private data
$100M: Arena said it reached a $100M annual revenue run rate eight months after launching its real-world AI evaluation product
21K: Oracle cuts 21,000 jobs as AI automates 13% of workforce. The company spent $1.8B on severance while cloud infrastructure revenue surged 93% and capex jumped 162% to $55.7B
5%: Sam Altman proposed a deal granting the US government 5% equity stake at OpenAI, forming public-private partnership
📰 What’s going on
Anthropic in talks with Samsung to create custom AI chip following OpenAI-Broadcom to align with industry shift towards bespoke silicon
Anthropic’s Fable is back after export controls were lifted. The relaunch adds a new cybersecurity classifier, a filter that catches risky requests before Fable answers. If a request gets flagged, users are routed to Opus 4.8 instead. Paid users can try Fable 5 through July 7 for up to 50% of weekly usage limits
Anthropic also launched Claude Sonnet 5; it delivers performance close to Opus 4.8 at a lower price point, with stronger reasoning, coding, and tool use than Sonnet 4.6, and improved cost efficiency at medium effort levels
OpenAI previewed GPT-5.6, a new model family led by Sol, its flagship next-generation model. The other tiers are Terra, a balanced everyday model, and Luna, the faster, cheaper option. The unusual part is the rollout. OpenAI said the U.S. government asked it to start with a trusted-partner preview before broader release
Google dropped an upgrade to its Nano Banana series of text-to-image models. The new Nano Banana 2 (NB2) Lite is the fastest and most cost-effective image generation option currently available from Google. It can produce fresh images in a mere 4 seconds, and costs just $0.034 per 1,000 outputs. Tradeoffs: NB2 Lite supports just 1k resolution, as opposed to the 2k and 4k images by the classic Nano Banana models
Gemini Spark, Google's agentic assistant, launches on Mac. It now works with local files, integrates with Keep, Tasks, and third-party apps like Instacart and OpenTable, and adds real-time topic tracking for AI Ultra subscribers
Cursor has officially launched its iOS app that lets you manage coding agents on your mobile. The app works in two modes: you can start entirely new cloud-based agents directly from iOS, or remotely control agents already running on your desktop
Meta has quietly launched its own new mobile app, Pocket, and it lets users create what it calls “Gizmos:” vibe-coded mini apps and games that users can publish to their social feeds
Meta introduces an AI subscription model that turns smart glasses into SaaS. Users can pay monthly fees to access advanced AI-powered features
Meta is developing a cloud infrastructure business that would sell access to AI compute power and models. The move would put the company in direct competition with AWS, Google Cloud, and Microsoft Azure as it seeks to monetize its massive data center investments
GitHub added AI credit pools to cost centers, giving Copilot admins a way to stop one team from draining shared monthly credits
Notion announced support for HTML inside documents, which allows creators to embed interactive widgets. The vision here is to allow teams to embed snippets of HTML that bring things like prototypes, explainers and other things to life inside a document. This collapses the gap between documentation and interaction. Instead of describing a feature flow in text or linking to a separate prototype, you could build lightweight, functional demos right where your specs live
Walmart, Uber and Microsoft are among the orgs capping AI usage as ROI scrutiny intensifies. Uber depleted its annual AI budget in months, and some procurement teams now demanding line-item accountability for every tool
X shipped an MCP server that connects Grok, Cursor, and Claude straight into the X API, so your agent can search the full post archive, check trends, manage bookmarks, and draft Articles using your own account’s permissions
X has begun rolling out X Money to select Premium+ users, introducing a cash sweep program offering up to US$10M in FDIC insurance. The service also includes a 6% APY, a Visa-backed debit card, p2p transfers, and cashback, with a broader rollout planned for mid-2026
📚 Good reads
Summer reads. Curated book recommendations by a16z and MIT. Nice selections including new and classics, encompassing science, entrepreneurship, fiction and culture
Where AI is heading next. Peter Yang expects that AI “super apps” like Codex, Claude, and Cursor will become the main workspace, turning many traditional SaaS products into invisible back-end pipes used mostly by agents. Always-on cloud agents and multiplayer workflow (where teams and agents collaborate across tools) are the next milestone
Grading AI fluency in Product orgs. Adam Fishman argues that AI “usage” is a vanity metric and that judgment is what matters. He shares a 4-tier rubric from his Mozilla PM org (unacceptable, good, better, best) that focuses on leverage and discernment. Ultimately, he says PMs who treat AI as a core skill, leaving a clear human fingerprint on every AI-assisted artifact, will advance faster
Understanding the code our AI agents write still matters. Geoffrey Litt explains that even as agents get better at generating and self‑verifying code, humans still need to understand that code to stay creative and in control. Instead of line‑by‑line diff reviews, he suggests higher‑leverage techniques like code explainer docs, quizzes to test your own mental model, or small “microworlds” you can poke at
How to write OKRs for an AI product. Jeff Gothelf shares why AI OKRs should avoid output metrics like “95% accuracy” and instead focus on observable user behavior. He proposes 3 types of KRs: outcome (what users do after the AI responds), calibration (how accuracy shows up in reduced checking/correcting), and trust (how much work users let the AI do for them)
Zapier CEO bans DMs aiming for transparency and AI context. Interesting approach by Zapier’s Wade Foster, basically banning exec DMs to push almost all conversations into public Slack channels, treating the company’s knowledge as a “Shared Brain” for AI. They track an exec “transparency leaderboard” (% of messages in public channels), and record and transcribe meetings. Too much?
🎧 Good listen/watch
[Podcast] PMing at Meta today. Meta PM leader Jagjit Chawla explains how they’ve blown up old-school PM rituals: PRDs are now a paragraph + prototype, status decks are replaced by an overnight agent that reads code, docs, and email, and reviews start at the decision. Ideas and first-principles judgment are the new bottlenecks
[Podcast] How Gusto built a product line with Claude Code, and putting GLM 5.2 to test. Claire Vo shows how Z.ai’s new GLM‑5.2 is strong enough to rival frontier models for many coding and agentic workflows, especially when cost and control matter. She walks through wiring it into Cursor and Claude Code, and stress-testing it. Separate, but on the same episode, Gusto’s Eddie Kim shares how a 5‑person team, a “permanent Zoom” with Claude Code, and almost zero traditional process shipped a tier‑one production agent in 10 weeks by treating AI as a team member
[Video] The future of work with Claude. Anthropic’s product and eng leads explain how Claude Tag lives inside tools like Slack, watching public channels and autonomously running long, multi-day workflows. Tag can self-schedule work and adapt to team preferences. For PMs, this is an always-on agent sitting in the middle of your team’s work, driving PRs, answering questions, and diffusing best practices across the org
🧑💻 Worth learning
Loop engineering. Marily Nika explains how to design loops so AI agents don’t fail, over-iterate, or hallucinate when given vague goals and unchecked tools. Designing loops is becoming a core PM skill in a world in which we delegate more and more to AI
39 principles for better human–AI interfaces. This piece turns academic work on human–AI interaction into a practical framework and 39 principles for building AI products users can trust. For PMs, it’s a checklist for when to delegate to AI, how much autonomy to grant, or how to keep assistance useful without becoming manipulative or unsafe
🔧 Products to try
Figma Motion puts animation timelines inside the Figma canvas, so designers can create keyframes, prompt for motion, and hand developers code-ready timing specs; free options
BrowserAct gives your agent a real browser for clicking, filling forms, uploading files, handling verification, and scraping live sites; free limited tier
OpenClaw app. The AI agent tool is now on iOS and Android; free to try
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!
