Your AI is only as good as what you feed it. Feed it a transcript with your product names, acronyms, and numbers spelled wrong, and you spend the afternoon hand-fixing the output.

Wispr Flow Notetaker uses your dictionary and calendar, so product names, acronyms, numbers, and uncommon names come out spelled right. It pops up when your meeting starts and captures Zoom, Google Meet, Teams, and Slack huddles with one click, no bot joining the call.

Then your meetings show up inside Claude or ChatGPT through the built-in connector. No copy paste. Try Notetaker free on Mac and Windows.

Try Notetaker Free

On Wednesday, Google shipped its second frontier model in a month. By Thursday morning, investors had wiped more than $500 billion off its market value. Microsoft, which has no frontier model at all, had the best day of the tech earnings season so far. The model wars are speeding up again, but the scoreboard that matters tracks who gets paid.

the best model, the worst day.

A few hours before its earnings call, Google announced Gemini 4 Carbon, two weeks after Argon went into general release. Google says Carbon ranks first on the Artificial Analysis Intelligence Index with a score of 59. It also leads the benchmarks for finding, exploiting and patching software vulnerabilities, and it can work for days on a single engineering problem, pushing through dead ends to trace bugs in large codebases. Google adds that Carbon cut successful cheating on Terminal-Bench to zero.

However, regular folks like you and me can’t use it just yet. A small group of cyber defenders gets access first, then trusted partners, while Google takes part in the U.S. government’s voluntary pre-release testing. Pricing will match Argon’s introductory rate of $2 per million input tokens and $10 per million output tokens. Google went nearly a year without a new flagship, so two in a month is a statement, and Pichai told analysts that more are coming.

Then the earnings came out. Search revenue grew at a mere 12%, down from 17% in the second quarter and 19% in the first. Google’s chief business officer said search queries are growing at the fastest rate in five years, driven by AI Mode, but those queries earn less than classic search. Capital spending reached $51.4 billion for the quarter. Free cash flow was negative for the second quarter in a row. Google’s future commitments for compute, energy and data centers now exceed $1 trillion. The stock fell about 13%, the biggest market-value drop in the company’s history.

The one bright spot makes the problem sharper. Google Cloud grew 103% to $30.7 billion, and some of its best customers are AI labs building assistants that compete with Google Search. Anthropic’s S-1, filed Monday, commits roughly $114.5 billion to Google for compute through 2033. Google is selling picks and shovels to the people digging under its own house.

The bull case deserves a hearing. The Journal’s Heard on the Street column calls Alphabet the most complete AI company there is. It has a frontier model, a fast-growing cloud, its own chips, an ad machine, $265 billion in cash, and a Gemini app that just passed a billion users. It still trades at about 20x forward earnings, against Microsoft’s 25x.

Still, both readings could simultaneously be correct. Benchmark leadership now changes hands every two weeks, while a business model takes years to shift. Google’s AI answers your question so well that you never scroll down and see the ads, below, that pay for it. The practical lesson for anyone choosing tools: no lab stays ahead for long, so pick tools by the task in front of you, not by loyalty to a lab.  But you knew that!

also on the radar.

Microsoft wins without an AI model.
Azure grew 48%, which lifted quarterly revenue 19% to $92.3 billion and profit to $36.6 billion. On Thursday, Microsoft began marketing at least $25 billion in bonds, its first U.S. high-grade sale since 2017, in up to ten tranches with maturities as long as 40 years. Microsoft and Johnson & Johnson are the only U.S. public companies rated AAA by both Moody’s and S&P. Hyperscalers have raised more than $313 billion in investment-grade bonds this year, almost three times the total for 2025. One detail got less attention than it deserves. Last quarter, Microsoft lowered its 2026 capex forecast after extending the estimated useful life of its data centers to 25 years from 15. The AI boom is becoming a credit story, and in a credit story, the accounting matters.

your phone is paying for the data center.
According to TechInsights, AI hyperscalers have pushed flash storage prices up 5x in a year, and they now buy twice as much flash as Apple. The Journal estimates Apple will spend $25 billion on flash this year, for roughly the same amount that cost $6 billion in 2025. The extra 256 gigabytes in an iPhone Pro cost Apple $15 last year and $66 now, and iPhone prices are up 13% on average. Tim Cook called it a “100-year flood.” Translation: you’re paying for the chatbot boom even if you never use a chatbot.

Doordash buys Instacart.
DoorDash agreed to buy Instacart for about $17 billion, or $65 a share in cash and stock, a 26% premium. The deal is squarely aimed at Walmart and Amazon. The piece DoorDash really wants is Instacart’s advertising and enterprise software business, which now makes up 30% of its revenue and is growing faster than delivery. Closing is expected in late 2027 at the earliest.  The deal also ends the exclusivity agreements that tied grocers to one delivery platform, which means more choice for grocers and a bigger rival for Walmart and Amazon.

Amazon sharpens its knives.

In a potentially aimed at increasingly popular AI shopping agents, Amazon has expanded its ShopDirect program to 250 million products, up from 100 million in March.  Shoppers can click through to the other retailer’s site. For some products, they can have Amazon’s agent complete the purchase without leaving Amazon, a feature called Buy For Me. The listings are also appearing in more places than search results and the Alexa shopping chatbot.  Here, Amazon is taking aim in particular at Meta’s Muse, which, you’ll remember, surpassed 10 million downloads in September, and which Amazon blocks from accessing its site.

worth your time.

Felix Rieseberg is an engineering manager at Anthropic who has led work on Claude.ai, Cowork and Claude Code Desktop. He writes that he stopped giving coding interviews this month because Claude now handles the small challenges better than candidates do. His conclusion is optimistic: building has become cheap, and newer engineers are passing veterans because of it. His advice is to find the goal your manager’s manager cares about, pick one radical idea, and build a working prototype over a weekend. His proof: two relatively new engineers prototyped a graphical interface for Claude Code in one weekend, and it shipped about eight days later. He now screens candidates for taste, empathy and principled thinking. No one has built a benchmark for those yet.The human mind is the original generative machine.