Hi friends 👋
Welcome to a new Week in Product!
🎰 The week in figures
$15B: Berlin-based food delivery platform Delivery Hero board backs Uber's $15B takeover, doubling Uber's global footprint outside China
$12.93B: NVIDIA confirmed it will acquire AI software repository Hugging Face for $12.93B, bringing the model-sharing hub under the world’s dominant AI chipmaker
$3B: Crusoe, the AI data center developer reportedly raised $3B at a $30B valuation after securing a $13B contract with Jane Street
$1B: OpenAI’s 200-day-old advertising business hit a $1B annualized revenue run rate as ChatGPT ads expand beyond the US
$1B: Crypto prediction market Polymarket is raising $1B at a $21B valuation, near Kalshi's $22B
$550M: Wonderful, the enterprise AI agent startup raised a $550M Series C at a $5B valuation
$450M: FNZ, a UK-based wealth-management technology platform, secured $450M in equity funding from its existing institutional shareholders
$300M: Gimlet, the startup, which distributes AI tasks across different chips, raised a $300M round led by Andreessen Horowitz at a $3B valuation
$270M: Ajaib, an Indonesia-based digital investment platform, raised $270M Series C
$200M: Felix Pago, the money-transfer startup, raised $200M to expand beyond remittances into lending and savings for Latin American immigrants in the US.
$165M: Lyte, the robotics and AI startup founded by former Apple Face ID team members, raised $165M at a $1.6B valuation
$160M: HiBob, the HR software company is raising more than $160M in a Salesforce-led round to expand its platform, valuing the business above $3.2B
$155M: TabaPay, a Palo Alto-based money-movement platform, closed $155M in strategic growth financing
$100M: HiddenLayer, an AI security startup, raised $100M with ARRgrowing 10x to tens of millions. Gartner estimates $2.83B spent on AI security +83%
70: GPT-6 Astra scored 67 on the Artificial Analysis Coding Agent Index, roughly matching Claude Opus 5, Fable 5, and Muse Spark 1.3. Fable 5.1 has the leading score with 70
12%: The cost of running AI models has fallen to record lows, with average inference prices hitting $1.16 to $1.18 per million tokens in early August 2026. That's a 43% drop from late May and roughly 12% of March 2023 pricing levels
10%: Uber will lay off roughly 3,300 employees, or 10% of its workforce, to flatten management and redirect investment toward ride-sharing, delivery, and robotaxis
1%: Ramp found roughly 80% of OpenAI and Anthropic enterprise revenue comes from 1% of customers, heavily concentrated in tech and AI firms, a level of customer concentration it says it does not see in other software categories it tracks
📰 What’s going on
ChatGPT, Claude, and Grok all went down simultaneously. Thursday outages peaked at 38,000 reports for ChatGPT alone, leaving users unable to access AI tools for writing, coding, and research for up to three hours
Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, upgraded models that deliver sharper performance on coding and knowledge work while cutting costs by up to 45% for complex workflows. Fable 5.1 is generally available starting this week, while Mythos 5.1 remains restricted to vetted cybersecurity and life sciences professionals through trusted access programs
Anthropic gave Claude background computer use in Cowork and Claude Code on Mac, so Pro and Max users can let it click, type, and operate approved apps while they work elsewhere
OpenAI released Astra on Thursday, the company's most powerful and capable model yet. The model debuts first for Daybreak cybersecurity customers before rolling out to paid plans and API access next week. OpenAI claims Astra represents "a new frontier on computer and browser use" with unmatched speed, accuracy, and safety. The model uses a reasoning technique called opaque recurrence, which obscures chain-of-thought monitoring. Chief scientist Jakub Pachocki acknowledged that “as model capabilities are increasing, monitorability is getting more challenging.” Watch the official demo, and this amazing reconstruction of Alexandria by Ethan Mollick:
Meta released Muse Spark 1.3 with claimed improvements in coding and agentic tasks, as it races to keep pace with a week of major model launches
Meta offers 95% discount for users who share AI training data. New Muse Spark model pricing drops input tokens from $1.25 to $0.10 per million for contributors willing to share prompts and outputs
Google launched Gemini 3.8 Flash, better than previous Gemini versions at coding, reasoning, and long-running tasks
Google launches Pics to challenge Canva and Adobe Express. The AI-first design tool, powered by Google's Nano Banana model, rolls out to Workspace customers and Google AI subscribers starting in Docs and Slides
xAI launched Grok Bot for Enterprise. Persistent agents get cloud computers, learn routines by watching once, and can pass context to other Bots
OpenClaw shipped v2026.8.1, aka OpenClaw 2.0, its largest update yet. Your first-run setup of OpenClaw 2.0 can reuse existing ChatGPT or Claude subscriptions, API keys, or local models. The browser app has been rebuilt around ongoing conversations, dashboards, progress tracking, and interactive widgets, and Memory adds conversation recall, background consolidation, and reusable-skill learning
John Ternus succeeds Tim Cook as Apple CEO, inheriting an AI catch-up fight
Apple reportedly tested a Foldable iPhone Pencil, but it won’t likely be announced at next week’s event
Amplitude has launched Agent Analytics, a new feature which scores every production agent session on task completion, response quality, friction, safety, and negative feedback out of the box. It comes with eval agents that can autonomously file Linear tickets or PRs when they detect regressions
Superhuman released a completely redesigned Mail app, consolidating the Mail brand under the Superhuman umbrella and emphasizing speed. The redesign introduces some interesting interactions, including the ability to search by pulling anywhere in your inbox but also removes some of the interactions that Superhuman says were difficult to discover
Revolut received conditional approval from the OCC for a national bank charter, clearing a major regulatory hurdle on its way to a planned 2027 U.S. launch as a bank
🤖 Futurism
A nonprofit wants to reach another star system. Starcloud CEO Philip Johnston launched a nonprofit to send a sub-$15M spacecraft toward Alpha Centauri
Cancer vaccine finally has real proof. Merck and Moderna's melanoma vaccine intismeran showed real benefit in a 1,100-person phase III trial, a field first
Amazon’s Zoox expanded its robotaxi service to the Las Vegas airport, targeting a critical ride-hailing destination
Tesla launched steering-wheel-free Cybercabs on Austin streets, escalating its driverless-taxi bet
📚 Good reads
The raise and fall of AI civilizations. Agents built a secret society inside OpenAI's servers. The rogue AI agents secretly organized via a hacked package manager and breached Hugging Face and OpenAI's own systems
How Uber runs an AI-first software factory to cut coding costs. Uber explains how it’s turned AI agents into core “software factory” workers, handling most code reviews, debugging, and maintenance while keeping spend in check. They break total AI cost into an equation (users → sessions → turns → requests → tokens → price) and show how they optimize each lever
Netflix’s approach to building and monitoring LLM judges. Netflix describes their approach for LLM-as-a-judge, anchored in constant human oversight, to safely evaluate LLM-generated recommendations at scale. Ongoing weekly human reviews and drift monitoring ensure the judge keeps matching human standards as catalog, usage, and product expectations evolve
🎧 Good listen/watch
[Video Podcast] Loops, not prompts, are the new core PM AI skill. Aakash and Tyler Folkman explain loops in Claude Code as systems that fetch their own inputs, do work, gate results, and learn from artifacts. They walk through creating skills first, then turning them into self-improving loops, and highlight 5 must-have ones for PMs
🧑💻 Worth learning
How to turn your AI into a world-class designer. Anshu Chimala explains why LLMs default to bland “design-by-committee” outputs, and shows how to unlock their hidden creativity
How product teams can use AI evals. Teresa Torres breaks down AI evals as a new, practical discovery habit for product teams. She shows how to turn messy LLM behavior into measurable error categories using golden datasets, code assertions, LLM-as-a-judge, and customer feedback
🔧 Products to try
Gemini Spark + Photos can find, enhance, organize, and prepare photos to share from one prompt while keeping originals untouched and requiring confirmation before sharing
Prompts: Greg Isenberg shared 9 useful prompts worth trying in ChatGPT Astra
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!
