A faster Claude, a voice model that can act, and a new kind of traffic jam at your favorite restaurant. Here is what moved today.
7 Stocks You May Never Need to Sell
Buying a stock is easy. Finding one you may never need to sell is harder.
This free report reveals seven blue-chip companies selected for financial strength, durable demand and long-term potential to hold up when markets turn ugly.
TODAY IN AI
3 things moving fast
1. Anthropic ships Claude Sonnet 5.5, 30% faster at the same price.
Sonnet 5.5 uses adaptive thinking by default and handles a context window of 1 million tokens. Pricing holds at $2 per million input tokens and $10 per million output tokens, the same as Sonnet 5, but Anthropic says the model typically uses fewer tokens, so comparable work can cost up to 30% less. Read Anthropic's best practices for building with it.
2. ElevenLabs launches Eleven v4, a voice model that reads tone and pacing.
The model interprets emotion and context, and it supports more than 90 languages. A Turbo version starts speaking in about 150 milliseconds in company testing, which matters for voice agents that have to answer in real time. ElevenLabs says v4 ranked first on the Artificial Analysis voice leaderboard for September and won about 75% of blind listener comparisons, though both figures come from the company. Unite.AI has the latency comparison.
3. xAI opens Team Bots, shared Grok assistants, in public beta.
One bot serves a whole team while keeping each person's conversations private. Teams feed it documents, brand guides and internal processes, connect tools such as Salesforce and GitHub, and the bot remembers corrections so answers improve for everyone. It is available on Teams and Enterprise plans only. Read xAI's announcement.
When Postgres analytics slow down, most teams add a second database. Then comes sync lag and drift. TimescaleDB extends Postgres instead: analytics on live data, no pipeline. Get $1000 to start.
FROM THE FRONTIER
AI agents that work perfectly can still break things
The table for two. Venture capitalist J.C. Bar-de Stefano asked a personal assistant called Instinct to keep hunting for a table at a New York steakhouse. The agent sent hundreds of requests per hour to Resy, the platform banned his account for bot activity, and access came back only after he contacted the company. He has since stopped using agents for bookings.
The fine print. Resy's rules bar unauthorized third-party bots and agents, so a helpful assistant looks like abuse from the platform's side. Restaurant owners quoted in the same report worry that reservations will turn into a digital race that ordinary diners lose.
The bank run. Apollo chief economist Torsten Slok floated a bigger version of the same problem. If assistants like Muse moved household cash out of checking accounts paying about 0.1% and into fintech accounts paying 3.3% to 5.0%, banks could lose the cheap deposits they lend against. Slok presents it as speculation, not a forecast.
The pattern. None of these stories needs an agent to malfunction. One assistant doing its job is fine, and millions doing it at once is a load test nobody planned. The likely response is more blocking and more rules about which agents may act on your behalf, so before you hand an agent your accounts, check whether the service allows it.
200+ Proven Ways to Make Money With AI in 2026
The next wave of millionaires will be people who figured out how to make AI work for them.
The window to get ahead is still open. But not for long.
Here are 200+ proven ways to make money with AI in 2026.
Sign up for Superhuman AI, the free daily newsletter read by 1M+ professionals, and get instant access to all 200+ ways to profit from AI this year.
IN THE KNOW
Turn the effort dial. Anthropic's developer guide says to start Sonnet 5.5 at medium effort for well-defined coding work and move to high for complex tasks. It also says to lower the effort setting instead of prompting the model to think less. The full guide is on the developer blog.
Leaderboard check. On the Artificial Analysis Intelligence Index, Opus 5.5 at max effort sits on top with a score of 58, while Sonnet 5.5 at max effort scores 56. See the current rankings.
A human backstop. Startup Wajo builds an agent called Fo that schedules appointments, arranges vendors, changes travel plans and makes purchases. It can also bring in human workers for tasks the AI cannot handle. Pricing is not listed yet.
Direct the voice. Eleven v4 lets you steer a performance with inline tags in the text, such as [laughs], so you can mark where a line should turn playful or sharp. ElevenLabs explains how the tags work.
PROMPT STATION
Turn any announcement into a voiceover script
This prompt works in Claude, ChatGPT or Gemini and turns a product update, press release or news item into a 30-second script written for the ear. It adds delivery directions in square brackets, the kind of tags a voice model such as Eleven v4 can act on. Try it on the last announcement you read, since new voice models make this worth testing this week.
Act as a voiceover director who writes for the ear, not the page.
ANNOUNCEMENT TO COVER: [PASTE ANNOUNCEMENT OR NEWS HERE]
AUDIENCE: [WHO WILL LISTEN]
TONE: [TONE, SUCH AS WARM AND DRY]
Write a 30-second voiceover script of about 75 words. Use short sentences and plain words. Add delivery directions in square brackets before the lines they affect, in the style of [laughs] or [said quietly]. Use no more than five directions. End with one clear next step for the listener.
Then give me a clean version with the brackets removed so I can compare how each one reads.
If any fact in the announcement is unclear, ask me before guessing.