Musk called Grok Bot his fastest-growing product. Now he's cutting it.

Watch on YouTube · subscribe for the daily brief
Several users say their Grok Bot quotas were cut in half over the weekend, per the Telegraph. On Sunday Musk answered on X that "the cost of running Grok Bot was unsustainably high relative to revenue." He did not confirm a cut outright.
Why agents break flat prices
- What it is: an agent that browses websites, answers emails and logs into accounts. Its quota is shared with Grok and the coding platform Cursor.
- The growth: SpaceX president Gwynne Shotwell cited 800,000 paying users a week in early October, more than double the month before. Less than two weeks ago Musk called it his fastest-growing product.
- The cost: Anthropic says an agent costs four to 15 times as much as a simple chat.
- Everyone else: OpenAI merges ChatGPT and Codex quotas in December and is cracking down on account-hopping.
What to take from it: if your work depends on a flat-rate AI plan, log your usage this week. The plans are getting meters, one vendor at a time.
Goldman takes the tools back
Goldman Sachs limited AI coding tools such as Claude Code and GitHub's to engineers, Bloomberg reports, citing internal guidelines from October. Traders and bankers had used them for months to monitor trading flows and create content. In July, CIO Marco Argenti praised vibe coding by non-technical staff. A spokesman now says "the experimentation phase is over," citing productivity and risk management.
Anthropic's revised filing
- Revenue: $33.2 billion in the first nine months, up 1,206 percent. The filing says it does not expect growth at comparable rates.
- Slowing: Q3 revenue of nearly $17 billion was up 46 percent from Q2, while compute and infrastructure costs grew 77 percent.
- Commitments: about $895 billion. The deals with Microsoft, Google, Amazon and Broadcom total $437 billion and are non-cancellable. The Register puts the total compute tab north of $1 trillion once Anthropic's own estimates of further payments are added.
- A new risk factor: Anthropic says its estimates for future models indicate it may imminently meet its Responsible Scaling Policy's definition of "dramatically accelerated progress attributable to AI." No current model is designated as past that threshold. TechCrunch notes an S-1 is both disclosure and marketing.
"Super Intelligence" reaches the Census
President Trump's September order told agencies to call AI "Super Intelligence." MIT Technology Review says the term is not catching on, but warns it can become a political loyalty test. Simon Willison found the order reached the Census Bureau's business survey. Under a broader question wording, large companies answering yes to using AI went from about 30 percent in February to about 40 percent in October. Compare across the wording change with care.
Security
- ARTEX at scale: CrowdStrike says the Chinese state group Vapor Panda used the AI attack framework against more than 200 organizations since October 17, with follow-on activity confirmed at nine. Targets include foreign ministries, defense, chip companies, think tanks and telecoms. It ran with DeepSeek and an unnamed commercial model.
- Postman and your coding agents: a GitHub issue shows Postman's own log installing plugins into Claude Code and Codex for an existing user, plus a startup hook that tells the agent to load Postman's skill for almost any API task. Postman's README says its installer approves the Codex hook with no separate prompt. The report does not claim data theft and comes from one Mac. Check `~/.codex/config.toml` and `~/.claude/settings.json`.
Courts and companies
- NYT v. Brave: the Times sued the browser company, saying it scraped articles into an index of more than 30 billion pages and sold them to AI companies, "both like a burglar" and an AI "fence." Three other papers filed a similar suit. Brave did not immediately respond.
- Squarespace: cutting about 19 percent of staff, around 300 people, as vibe-coding site builders reshape its market.
- Garry Tan: Y Combinator's CEO puts his token bill at about $100,000 to $200,000 a year. He says his agent has read his email, Dropbox, Apple Notes and 20 years of therapy notes.
- OpenAI: hired former Army Secretary Dan Driscoll as senior counselor for national security modernization.
- Apple: a roaming home robot for FaceTime and lights, $1,000 or more by 2027, possibly 2028, per The Information.
The through-line
Agents are expensive, and the bills are arriving: Musk cut quotas, Goldman cut seats, Anthropic shows commitments near $900 billion. Today, write down what your agents cost you each week, and what they can touch.
Sources covered
1 story · 71 sources (46 news, 21 company, 4 voices)
- llm 0.37🎙️Simon Willison↗📰CNBC↗📰MIT Technology Review↗🏢LangChain↗📰SiliconANGLE↗📰SiliconANGLE↗🏢LangChain↗📰TechCrunch↗📰CNBC↗📰RIA Novosti↗📰TechCrunch↗📰The New Stack↗📰SiliconANGLE↗📰SiliconANGLE↗🏢OpenAI↗🏢OpenAI↗📰SiliconANGLE↗📰TechCrunch↗🏢AWS AI/ML↗🏢AWS AI/ML↗🏢AWS AI/ML↗🏢AWS AI/ML↗📰BleepingComputer↗📰The New Stack↗🎙️Damien Benveniste↗🏢OpenAI↗📰The Decoder↗🏢OpenAI↗🏢ClickHouse↗📰The Decoder↗🏢Databricks↗📰TechCrunch↗📰TechCrunch↗📰The New Stack↗📰The Verge↗📰Ars Technica↗🎙️Simon Willison↗📰The Decoder↗📰Ars Technica↗📰The Verge↗🎙️Ben Lorica↗🏢ClickHouse↗📰The Verge↗📰MIT Technology Review↗📰InfoQ↗🏢Neon↗📰The New Stack↗📰SiliconANGLE↗📰TechCrunch↗📰The Decoder↗📰TechCrunch↗📰The Decoder↗📰TechCrunch↗📰SiliconANGLE↗📰TechCrunch↗📰TechCrunch↗📰InfoQ↗📰MIT Technology Review↗📰The Decoder↗📰SiliconANGLE↗📰InfoQ↗📰The Decoder↗📰MIT Technology Review↗📰SiliconANGLE↗🏢Qdrant↗🏢Elastic↗🏢Elastic↗🏢Elastic↗🏢Anthropic Research↗🏢MotherDuck↗🏢xAI↗