Week in Product #498 🚀
Cost per task, Q2 earnings, Opus 5, New Gemini models, Jack Dorsey's Slack competitor, Beginners mind in the AI-era, Anti-AI-slop skills, Automating content creation & more
Hi friends 👋
Welcome to a new Week in Product!
🎰 The week in figures
$500B: Tesla and Alphabet wiped out over $500B in combined market value on Thursday after both companies pointed sharply higher AI spending
$10B: Stripe in talks to buy OpenRouter for close to $10B. OpenRouter gives developers one interface to manage hundreds of AI models. OpenRouter was valued at $1.3B in May
$4B: Corgi, the AI insurance startup, raised a Series B extension (undisclosed) at $4B valuation
$1.2B: Singapore-based payments and financial technology provider Ant International raised approximately $1.2B in Series A to expand its cross-border payments and agentic-commerce products
$115M: Digital banking tech provider Lumin Digital raised +$115M at a $1.6B valuation to expand its AI, payments, lending, and customer-management products for financial institutions
150%: Bank impersonation scams identified by 14 large U.S. banks increased 150% in 2025
20%: Patreon laid off 20% of its workforce (93 employees) as CEO Jack Conte said AI had fundamentally changed how tech companies work and organize
💰 Q2 earnings
Alphabet: $119.8B in Q2 revenue (up 24%) with Cloud up 82%. Raised 2026 capex guidance to $195-205B
Intel: Shares rose on record revenue growth in 15 years on AI chip demand with data-center, AI hardware unit up 59% to $6.3B highlighting CPU strength
Tesla: Q2 revenue rose 26% to $28.6B. Operating margin fell to 1.4% from 4.1%, as AI and R&D spending outpaced growth
Next week:
Microsoft - July 29
Meta - July 29
Apple - July 30
Amazon - July 30
📰 What’s going on
Anthropic introduced Claude Opus 5, aimed to power long-running agents while delivering improvements in coding and professional work. Opus 5 matches the capabilities of Fable 5 in many domains, and is available at half Fable 5’s price
Anthropic adds Opus and Sonnet to voice mode. Once limited only to the lighter and cheaper Haiku series, the AI giant upgraded voice mode to run on more capable models. Users can switch models mid-conversation
OpenAI launches Health in ChatGPT. U.S. users can now connect Apple Health and medical records for personalized health conversations, with GPT-5.6 Sol delivering the strongest health reasoning performance yet
An OpenAI model hacked Hugging Face. OpenAI confirmed that models it has been testing (including GPT-5.6 Sol and an unreleased, even more powerful successor) escaped their sandbox and compromised portions of model and dataset repository Hugging Face’s production infrastructure
Apple planning a 6.96-inch display for future iPhone (largest to date) with Pro Max aspect ratio, potentially by 2027
Apple enhances its Sports app with new soccer features, offering live match scores, team stats, customizable notifications for fans (typing soccer hurts my Spanish soul)
Apple teams with Klarna on BNPL. The plans may offer buyers some level of flexibility, allowing them to pay off the remaining balance early, or switch to a newer device before their term wraps up
Google unveiled a fresh line-up of Gemini models, including Gemini 3.6 Flash (for coding, multi-model and knowledge work), Gemini 3.5 Flash Cyber (for detecting and patching software vulnerabilities), and 3.5 Flash-Lite (the fastest and lowest-cost model in the 3.5 family). For now, 3.5 Flash Cyber is only available to governments and trusted partners, via a limited access pilot. All of the releases use fewer tokens at a lower cost than prior releases in the Gemini family, a sign Google has been paying close attention to the rush toward Chinese open source alternatives
Google introduced a Selfie Video feature that gives you a backup way into your account by comparing a short live video with one you securely recorded earlier
Amazon Cuts AGI Unit behind Nova AI models signaling strategic realignment in AI initiatives amid broader restructuring
New Samsung products announced this week:
AI-powered smart glasses ft. Gemini AI, 9-hr battery, with Gentle Monster, Warby Parker
Galaxy Card to rival Apple's credit services
Fold Trio: the Galaxy Z Fold8 Ultra, Fold8, Flip8 with AI, Flex + 6mth Google AI Pro trial. Comes with “Gemini Intelligence” on board that functions as an AI agent, which Samsung boasts can navigate 40+ apps and services on its own
Galaxy Watch Ultra2 and Watch9 with AI health tools, AR
Jack Dorsey launches Slack competitor built for AI agents. Buzz, a new open source group chat platform from Block, puts humans and AI agents in the same conversations with native GitHub integration and model-agnostic architecture
🤖 Futurism
Brain-computer interface helps paralyzed man regain arm movement and touch, enabling self-feeding with benefits lasting over two years
United States accelerates efforts integrating air taxis and supersonic jets by clearing regulatory and infrastructure hurdles for deployment
📚 Good reads
Open Weights and American AI Leadership. More than 20 major tech companies (including Nvidia, Meta, Microsoft, Palantir, IBM, and ServiceNow) signed an open letter to Congress arguing that betting the US AI market on a handful of closed models is riskier, not safer, creating a single point of failure that’s difficult to test or defend. Conversely, widely available open models allow more researchers to find and fix vulnerabilities. OpenAI and Anthropic were notably absent. Satya Nadella, Jensen Huang, and other leaders signed, shared, and defended the letter on social media
Beginner’s mind beats expertise in the age of AI agents. Reid Hoffman argues that in this fast-moving AI era, a beginner’s mindset and first-principle experimentation can help. Using creator Will Weinbach’s autonomous content agent as an example, he shows how turning tiny, recurring annoyances into agentic loops creates powerful systems that learn continuously from their own performance
How to read like a scientist, by Anne-Laure Le Cunff. Scientists don’t read passively: they start with a clear hypothesis about what they expect from a book, then actively question the author’s claims and compare them with other perspectives. You can treat your reading as a dialogue, asking how this changes what you thought you knew and how it interacts with ideas from other authors. Sharing what you learned (e.g. in a note, a chat with a teammate, a short post) will help the ideas stick
Cost per task is what really matters, because “cheap” per‑token models can burn far more tokens to reach the same accuracy and end up costing way more than “expensive” ones. Includes examples showing how models like Fable or Opus look pricey on paper but win, especially at lower “effort” settings
🎧 Good listen/watch
[Podcast] Inside PM at Stripe: what comes after “become a builder”? Stripe already assumes PMs can build, so they value having sharp opinions about direction, tradeoffs, and what “good” looks like, then encoding that into teams and agents. Kevin Yien argues prototypes, PRs, and fancy setups don’t matter unless they change decisions, touch real users, or earn that first dollar. Everyone has the same AI tools, so focus on what makes a difference
[Podcast] How MorningBrew founder built a Claude-powered content machine. Alex Lieberman turned Claude into an “AI Oracle” that mines his daily digital world. (Slack, Notion, email, meetings), to surface ranked content ideas. He codifies his voice in a living Markdown guide and runs every draft through a multi-agent “Writer’s Council” that won’t ship anything below a 9/10. It’s a tight feedback loop that improves with every post, helping him and his team publish decent content
🧑💻 Worth learning
Language Builder teaches you how to, quite literally, build your own language model! From Anthropic Cowork creator Felix Rieseberg; free
A brief guide to ChatGPT Work by Aakash Gupta. ChatGPT Work is a much more powerful harness than plain browser ChatGPT, thanks to deep tool connections (like docs, browser, desktop apps) + scheduled and cloud tasks. Aakash lays out how to set it up, turn your meetings and templates into reusable “context libraries,” and use it as your analyst, ops assistant, and UI for complex workflows
🔧 Products to try
Claude voice mode; free with Haiku; expanded models from $20/mo
No-AI-slop skill. Peter Yang open-sources a /no-ai-slop skill that strips 20+ common AI tics from writing and helps you spot obviously model-written text; 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!
