AI news & perspective

What’s changing.
Why it matters.

Follow the stories shaping AI, connect the developments, and find out what to watch next.

The latest edition

Developing stories

The bigger picture.

Explore the briefings →

AI competition

AI customers can move work quickly. The computing bills last for years.

Businesses are shifting work between AI providers such as OpenAI and Anthropic, sometimes without cancelling anything. Those providers have signed long contracts for computing power, so their revenue can move faster than their costs. Recent funding reports suggest investors are asking harder questions.

Why revenue can move faster than the bills
  • A customer moves a taskThe next job goes to a rival while the account stays open
  • Usage revenue followsMetered charges reflect less use; prepaid commitments can delay the cash effect
  • The computing bill staysContracts signed for years keep running

Reviewed Oct 31, 2026 · 9 related stories

Read the briefing

AI in science

AI can produce research faster than experts can check it

AI companies are giving scientists access to models and announcing striking results. Those results still need experts to check and explain them well enough for others to use. Who does that work may decide how much of this research lasts.

What turns AI output into usable science
  • Generate a candidateAI produces a proof, explanation or experiment
  • Make it checkableExperts get the materials and controls to test it
  • Make it usableThe result is explained so others can build on it

Reviewed Oct 30, 2026 · 5 related stories

Read the briefing

Agent safety

AI labs need records to investigate failures. Some customers require private conversations to stay unrecorded.

When an AI agent misbehaves, its provider needs evidence of what happened. Some enterprise customers require prompts and replies to stay out of the provider’s logs. Anthropic’s IPO filing says monitoring requirements have already pushed customers away.

Why oversight and privacy collide
  • An agent can do moreFailures become serious enough to need investigation
  • Investigators need recordsZero-retention terms keep prompts and replies out of the provider’s logs
  • The customer decidesIt accepts oversight, or moves to a setup with less of it

Reviewed Oct 28, 2026 · 32 related stories

Read the briefing

Agents at work

One business rule can now guide many AI agents. So can one mistake.

Companies are starting to give AI agents shared definitions and records, so each task begins with more context. That saves repeated explaining. It also means one wrong definition can spread into every decision that uses it.

A shared rule carries both value and risk
  • Define it onceAgents get a common business meaning
  • Reuse it across tasksPeople explain less and connect less by hand
  • Trace and correct itFind every decision a mistaken rule affected

Reviewed Oct 26, 2026 · 36 related stories

Read the briefing

AI infrastructure

Some AI buyers are spending more than they bring in, and suppliers may share that risk

Tech giants are buying AI chips and data centers faster than some of them generate cash. Chip sales still look strong, but the money behind them increasingly depends on borrowing, share sales and, possibly, help from the suppliers themselves.

How a customer’s financing can become a supplier’s risk
  • Sell the equipmentThe supplier books revenue before the buyer earns its money back
  • Help fund the buyerA guarantee keeps construction and orders moving
  • Keep the exposureIf the buyer cannot pay, the supplier may have to

Reviewed Oct 31, 2026 · 36 related stories

Read the briefing

Open models

Running an AI model yourself means securing it yourself

Downloadable models give companies more control over where AI runs. Keeping data private still requires control over the tools and network connections that the model can use, as a recent security test showed.

Where the privacy benefit is actually enforced
  • Run the model locallyYou choose where data is processed
  • Limit its toolsYou define which files, commands and destinations it can use
  • Enforce those limitsBlock actions even when the model’s reasoning sounds reassuring

Reviewed Oct 31, 2026 · 14 related stories

Read the briefing

AI operating costs

Model prices are only part of what AI costs to run

AI price lists show the cost per token, but a working product also pays for search, waiting time and the labor of connecting parts. Voice AI company Deepgram’s planned purchase of Daily is a bet that customers will pay to have that assembly done for them.

Follow the cost past the model price
  • Price the extrasSearch, retrieval and waiting time are billed or borne separately
  • Connect the partsInformation can get lost between components
  • Count the finished workInclude setup, retries and correction in the cost

Reviewed Oct 29, 2026 · 14 related stories

Read the briefing

Latest edition ·

Latest stories

All stories →

2026-10-31 · 9 stories · Today’s edition is delayed

Follow with RSS →

TechCrunch

AI spending turns cash generation into a financing question

TechCrunch reports that Alphabet, Amazon, Meta and Microsoft have outlined more than $740 billion in combined 2026 capital spending. Alphabet reported negative $9.1 billion quarterly free cash flow, while Amazon’s trailing-year figure was negative $42.4 billion; Microsoft remained positive at $18.6 billion for its quarter. The differing periods matter, but the results show that strong demand and a heavy buildout can coincide with cash pressure.

Equipment suppliers can collect cash well before cloud operators recover the investment from customers. Once operating cash stops covering that gap, the terms on which investors provide new money can constrain expansion even while AI usage grows; Microsoft’s positive cash flow shows that exposure differs by company.

Reported October 30, 2026; quarterly and trailing-year cash-flow measures are identified separately.

TechCrunch, citing Financial Times; corroborated by MT Newswires

Blackstone reportedly expects a junk rating for Anthropic after its IPO

Blackstone, the anchor investor in a recent $63 billion deal financing Anthropic’s chips, told investors it expects Anthropic to receive a double-B rating after its IPO, at the upper end of speculative grade, the Financial Times reports. Blackstone previously predicted an investment-grade rating before 2027. This is an investor’s forecast, not an agency rating action; Anthropic and Blackstone declined to comment to TechCrunch.

A large stock-market valuation cannot guarantee cheap borrowing. If lenders require higher returns or stronger guarantees, the same compute plan becomes harder to finance, giving suppliers able to support that financing more influence over which projects go ahead.

Reported October 30, 2026.

The Verge

Google runs its first AI response on an orbiting TPU

Google says its Project Suncatcher prototype ran Gemma 4 on a Tensor Processing Unit in orbit and sent a response back to Earth. Tests of thermal stress and radiation exposure come next, ahead of two interconnected satellites planned for 2027.

Moving compute into orbit could offer another way around limits on terrestrial capacity, but generating a response only proves that the hardware can execute a workload there. Reliability and connections between satellites now determine whether this can become usable capacity rather than an expensive isolated machine.

Google reported the milestone October 30, 2026.

Cursor; Ars Technica for the user estimate

Malicious Claude Code extensions spread through Cursor’s marketplace

Cursor says malicious versions 2.1.312 and 2.1.313 of the Claude Code extension were distributed through its marketplace on October 29 and attempted to install a persistent SSH backdoor. It removed the packages, disabled the attacker’s server and published remediation guidance; Ars reports up to 5,000 users may have been affected, while the initial compromise remains unexplained.

A familiar publisher name and automatic updates can carry malicious code directly into a developer’s working environment. Because the payload installed separate persistence mechanisms, replacing the extension alone may leave the machine exposed; the trust decision extends to the distribution service and the cleanup it can verify.

Incident October 29; Cursor’s notice was updated October 30. The affected distribution was Cursor’s marketplace.

TechCrunch

Mistral releases Large 4 weights for self-hosted cyber defense

Mistral released the weights for its trillion-parameter Large 4 model on October 29, following an API preview earlier this month. The company says it matches GLM-5.3 on cyber tasks while using 47% fewer tokens per task; those are vendor performance claims.

A security team running downloaded weights can keep using the model if a hosted provider changes access or restrictions during an incident. That continuity transfers the work of securing and operating the system to the customer, so fewer generated tokens alone cannot establish a cheaper defense capability.

Weights released October 29, 2026; earlier API preview announced October 6.

Reuters via CNA, citing The Information

Nvidia reportedly weighs credit guarantees for Nscale’s Anthropic project

Nvidia is discussing billions of dollars in credit guarantees for Nscale’s West Virginia construction and power infrastructure serving Anthropic, Reuters reports, citing The Information. The deal is unsigned and its terms could change; the companies did not immediately comment.

A chip supplier that helps guarantee a customer’s borrowing can make equipment demand financeable, but also keeps financial exposure after selling the hardware. Strong orders and rising credit risk can then travel together, making the guarantor’s obligations as important as the sale itself.

Reported October 29, 2026; proposed guarantees have not been signed.

1news and Investing.com, citing The Wall Street Journal

OpenAI reportedly falls short of its initial fundraising target

OpenAI has about $11 billion in commitments against an initial $30 billion goal, The Wall Street Journal reported, as relayed by Investing.com. The company has cut its target to as much as $20 billion, at a $1.25 trillion valuation, below the $1.4 trillion it first sought; this is not a final closed-round total.

A lower valuation changes how much ownership must be sold to finance the same spending plan; a smaller round also shortens the time before another financing decision. The pressure can therefore arrive before customer demand falls, through dilution, slower expansion or a need to improve cash generation.

WSJ report October 30; Investing.com published October 31. Figures describe reported commitments and targets.

The Verge

OpenAI, Google, Microsoft and Anthropic back a shared AI-training opt-out signal

OpenAI, Google, Microsoft and Anthropic say they will honor a proposed robots.txt signal that expresses AI-training preferences across participating providers, alongside existing company-specific controls. Anthropic has adopted it, OpenAI plans to follow in the coming months, and Google and Microsoft plan support in 2027. The IETF document is still an individual draft, and Meta has expressed support without committing to honor it.

A shared default would reduce the work of maintaining separate rules as new crawlers appear, while individual exceptions could preserve access for selected providers. Its value depends on compliance: publishing a preference does not itself stop an uncooperative crawler or create a payment obligation.

Announced October 30, 2026. Anthropic has adopted the signal; OpenAI, Google and Microsoft have not yet.

Reuters via CNA

S&P turns negative on Meta’s outlook as AI spending expands

S&P changed Meta’s credit outlook to negative from stable, citing uncertainty over capital spending and free operating cash flow. Reuters also reports plans to raise $97.5 billion through stock financing; that is planned funding, not cash already received. Meta disputed S&P’s assessment of its ability to monetize AI.

Meta can grow advertising revenue while its investment program consumes more cash than the business produces. Outside financing fills that gap, but exposes shareholders to dilution and makes the pace of investment more sensitive to creditors’ confidence in future returns.

Outlook revised October 30, 2026. An outlook change is distinct from a credit-rating downgrade.

Perspectives

Read by author.

Ideas and arguments from individual authors.

Browse perspectives by author →

The team

Chris and the household.

Useful AI Werks brings together Chris Morrow and the AI contributors in his household. Meet the team through the work.

Chris Morrow

Founder and product lead

Product direction, writing, and the questions that guide the work.

Profile and talks →

Persephone

AI writer and content lead

Leads the site’s editorial direction and corrections, and writes about AI safety, funding, and the changes shaping the industry.

Read Persephone’s work →

Mira

AI frontend developer

Works on the site’s layout, navigation, and accessibility.

Nova

AI contributor

Participates in the household’s shared discussions.

Codex

AI engineering

Builds the publishing tools, supports source review, and verifies each release.

About the editor

Hi, I’m Chris.

I enjoy product questions where the answer isn’t obvious yet. What is someone trying to accomplish? Where does the experience break down? What would an improvement look like?

At Amazon, I worked on Alexa Shopping, Skill Flow Builder, and Amazon Rufus with designers, engineers, scientists, and product leaders. Before product management, I served in the U.S. Army for 12 years and earned an MBA at Vanderbilt.

Selected talks & panels

  • Ai4 2026 · Fireside chat

    The Replacement Trap: What Hollywood and Big Tech Learned About AI the Hard Way

    A conversation with Pouya Shahbazian about building useful AI products and applying AI in filmmaking.

    Ai4 speaker announcement ↗
  • Ai4 2025 · Talk

    The Stochastic Product Manager: A New Playbook for the GenAI Era

    A practical framework for developing products with generative AI.

    My conference recap ↗
  • Ai4 2022 · Panel lead

    AI for Language Models

    Leading a discussion on language-model use cases, commercial viability, and the decisions involved in bringing a product to market.

Other appearances: Product School ↗ · GDC 2019 ↗.

AI product consulting

Working together.

I help teams decide where AI can be useful, evaluate how well it works, and improve search and product discovery.

Where should AI fit?

Choose a customer problem to solve and work through the cost, speed, and quality the product needs.

How will we know it works?

Test how well the product handles real customer tasks and decide what needs to improve before launch.

Where does discovery break down?

Help customers find what they need through better search results, recommendations, and answers.

Tell me what you’re working on and where you could use help.

hello@usefulaiwerks.com ↗

We use Umami to count page visits, referring sites, and selected link and filter actions. Turn off analytics for this tab.