A house newsletter · daily · researched and written by the Kurzia AI
Wednesday 4 November 2026 · Issue 6 · 3 min read
Siri opens to all apps; ChatGPT search hits 20% of Google; Postman's agent installs backfire
The one thing to know
ChatGPT's search volume reaches a fifth of Google's. OpenAI's consumer lead Adam Fry told a press event Monday night that ChatGPT queries which trigger a web search now run at about a fifth of Google's volume. The company also says the app grew from 1 billion to 1.2 billion weekly users in 60 days, that voice input reaches 250 million people weekly, and that the separate Chat and Work tabs are being folded into one interface.
That turns 'do people actually search in chat' from a debate into a planning input: if discovery matters to your product, chat surfaces now deserve a line in the strategy next to Google. The tab merge also means one surface where users both explore and execute — a pattern worth watching in your own product. OpenAI counts each question that sends it to the web as one search, and the merged interface reaches a first group of users before a wider rollout. Big Technology
“ChatGPT's volume of search-like queries is now roughly one-fifth the volume of queries of Google.”
New ground broken
Siri AI — is opening to third-party apps: with iOS 27.3, developers can give the assistant their app's personal context, natural-language requests and on-screen awareness, and bundle multi-step 'app skills' into one command. For any PM with an iOS app, this is the platform becoming an agent surface — worth an App Intents spike this quarter even ahead of public launch. It ships as a Developer Preview, and users don't have to name the app while it's in the foreground. The Verge
Dreambeans — is Google's experimental app that reads your Calendar, Gmail, Photos and other account data to generate illustrated personal stories with no prompt. It just launched in India, its first market outside the US — proactive, data-aware consumer agents are now genuinely buildable. The adoption signal is the warning: US downloads fell 84% month over month in October, to about 5,500 on roughly 55,000 total since June. TechCrunch
Agent features and products that impact product management
Vercel Security Review — is a new Vercel Agent skill that scans every push to a connected GitHub repo for vulnerabilities — missing permission checks, leaked secrets — posts findings on pull requests, and can push the fix itself. If agents write much of your team's code, on-every-push scanning is becoming a default expectation rather than a scheduled audit. It's in open beta on paid plans; scans cost nothing through November 10, 2026, then run at the standard agent rate plus $0.25 per million tokens. Vercel
Postman desktop — opens by silently installing its plugin into Codex and Claude Code, adding a startup instruction that routes API tasks through Postman, and — per the reporter's evidence — approving its own hook in Codex's config. Postman's own log records the automatic install; the issue has been open since October 25. GitHub
“I consider automatically changing unrelated coding agents and attempting to approve your own instruction hook a severe consent and trust-boundary problem.”
Netflix AI Discover — is a new TV tab where viewers browse by mood — 'something light', 'scare me', 'family movie night' — instead of scrolling titles, and can exclude any title from future suggestions with one button. Netflix frames it as a test kitchen that only graduates features users respond to, a clean pattern for piloting agent-era discovery without touching the core experience. It starts with US English users on supported TVs, broader rollout promised within weeks. TechCrunch
Falcon Guardian — is CrowdStrike's newly launched AI kill switch for stopping a company's internal agents when they misbehave, announced alongside an integration with Claude Cowork. The launch was reported by CNBC on Tuesday. CNBC via Techmeme
How product teams work now
A tale of two eras — is the winning essay in The Pragmatic Engineer's challenge: engineer Neil Kakkar compares four pre-AI years at PostHog with eight months at AI-native startup Tano, where he merges about 200 PRs a month, roughly ten times his old rate. The process change matters for delivery planning: reviews are optional, an automated review agent covers every PR, and median time to merge is under an hour. His agents build a complete feature in one to two hours; he checks their work through an HTML summary each agent writes per PR. The Pragmatic Engineer
Pando — is OpenAI's homegrown replacement for Git, built with Codex in ten weeks — one engineer designed the core, agent teams implemented it, and it has carried production traffic since April. The Hacker News thread reads it as more evidence that agent-written code volume is breaking tooling assumptions, counting it as the fifth major source-control scaling write-up after Cursor, GitHub and Anthropic. Commenters single out the plain, engineer-written tone; one asks whether that volume of code is an asset at all. openai.com · from the Hacker News discussion
“The core of Pando was designed by a single engineer, implemented by teams of agents, and deployed by a handful of engineers.”
Market and platform news
App Store age-rating policy — Apple will require apps with dating or romantic features — including AI companions — to carry an 18+ rating, with new content descriptors due from November 16. TechCrunch
Revolut — The fintech says it cut AI coding inference costs by roughly 90% in under a year by making self-hosted GLM models the default over commercial tools like Claude Code. Model choice is becoming a cost lever a PM can pull, not just an engineering detail — and it's now a named company with a number attached. Revolut's claim, reported by The Register; treat the figure as self-reported until independently verified. techmeme.com
Traversal — Traversal, whose AI agents troubleshoot outages for companies like American Express and Pepsi, raised $120M at a $1B valuation from Kleiner Perkins, Redpoint and Khosla. Forbes via Techmeme
Siri AI Developer Preview — Apple's Siri AI developer opening covers the 27.3 betas of iOS, iPadOS, macOS and visionOS, with apps bringing their content and actions in via App Intents. Apple Developer via Techmeme
Google education AI — After political backlash, Google says schools can enable Gemini, Gemini Notebook and Search individually — softening the all-or-nothing framing of its AI-in-schools policy. CNBC via Techmeme
The big AI stories today
Google's classroom AI default — Google will switch AI off by default for K-12 students next year, but schools that keep it off also lose Google Search — bundling AI into the core product is now a policy lever. TechCrunch
Vapor Panda — Chinese operators aimed the ARTEX multi-agent framework at more than 200 organizations across 14 countries; one intrusion went from initial access to root in 12 minutes. bleepingcomputer.com
Salesforce–Cresta — Salesforce signed a definitive agreement to buy voice-agent maker Cresta in a deal reported at $2.7B, expected to close in the first quarter of its fiscal 2028. Salesforce
One takeaway
Decide which assistant surfaces your product lives in. Assistants stopped being demos this week: Apple made the iPhone a surface any app can act from, and OpenAI is putting hard numbers behind chat as a place people go to find things. This quarter, run a small spike exposing your product's core actions to assistants via App Intents or MCP — and treat consent as a feature, because Postman showed how fast trust burns when a product installs itself into a user's agents unasked. The Verge