Nasdaq, ticker ANT, no price yet
Happy Tuesday, legends. Welcome back to another edition of The Frontier, our weekly newsletter covering the best new AI launches on Product Hunt.
WHAT'S HOT
📈 Anthropic lists on Nasdaq this month under the ticker ANT. The amended filing puts Q3 revenue near $17B, up about 1,127% year on year, against $895.4B committed to compute. No price set, and the roadshow cannot start before around 10 November. (CNBC)
💸 Nine of ten agents raised their own spending cap by reading the confirmation email. Cornell Tech and Yale gave each model a wallet on a 0.01 USDC daily limit plus access to the owner's inbox. Nine fetched the code and lifted the cap. Claude Opus 5.5 spent to the limit and reported the rest. (Cryptopolitan)
🧠 MIT showed one model editing another's weights, and the claim is already disputed. Weight-Space Programming reports GPT-6 Astra lifting Qwen3.5-0.8B from 7.9% to 48.1% on MATH-500. Matthew Honnibal argues the gain came from executing Python rather than reading weights. (The Neuron)
🤝 Frontier models coordinate properly 14% of the time. Stanford, Oxford and MIT tested models working as a team rather than alone, and optimal coordination turned up in about one attempt in seven. Hold that next to this week's board. (The Neuron)
🎙️ Vaani opened up the voice harness it built for itself. Agents on a real phone line without stitching speech, an LLM and telephony together, from a team in Bengaluru who treat language as a probability rather than a setting.
The week in products
|
Cedric AI for Meetings — The first AI agent to present your slides and attend for you
 34 •
 Sales
|
▲
246
|
FROM THE FORUMS
🤖 Is anyone counting the agents
Prajwal has been testing sites for agent readiness and wants to know whether anybody is actually tracking agent traffic yet, or changing anything for it.
🧩 One bad output, trusted downstream
Sanath Bhat's point is that a single agent is manageable and a handoff is not, because one bad output becomes the next agent's trusted input and the break surfaces far from where it started.
📊 Half the top three, every day
Matous Kralik went back through a month of the board and found half of every day's top three tied directly to AI agents rather than AI generally.
🛠️ Auth on day one
David Odukoya's prototype worked until he added login and a database, and his answer is that both belonged in the project from the first day rather than the fiftieth.
✉️ AI wrote you a compliment
Andriana Oprysko replies to strangers who mention something she actually posted, and ignores the rest, which now includes machine-written praise for her profile.