Live coverage of the September 9 event reported that Apple Foundation, the model layer behind Siri AI, is powered by Google, with a Gemini partnership underneath the year's biggest consumer agent launch.
Live coverage of the September 9 event reported that Apple Foundation, the model layer behind Siri AI, is powered by Google. The most visible consumer agent of the year would then depend on a rival's frontier model.
OpenAI's GPT-6 Astra, now rolling out across ChatGPT paid tiers, the API, Azure, and Bedrock, posts 98%1 on FrontierMath Tier 4, 99.9%2 on ARC-AGI-3, and 100%3 on ExploitBench. Reported cost runs roughly 31%4 below Claude Fable 5.1 on comparable work.
A weekly operator deep dive, the full Dojo Pro prompt packs, and the complete prompt database. Founding members lock the launch price forever.
Join the founding list →At its September 9 event, the first under new CEO John Ternus, Apple announced iPhone 18 Pro and Pro Max on the A20 Pro chip, its first folding phone, the iPhone Duo, and AirPods 5. The headline software is Siri AI, an agent built into iOS 27 that reads personal context across messages, mail, photos, and app...
Continue readingIran said on Tuesday it attacked 10 ships near the Strait of Hormuz after the US sank five Iranian oil tankers, the largest declared exchange of strikes on shipping in the six-month-old war, and also fired ballistic missiles at a base used by US forces in Jordan.
US stocks fell for a third straight session, the Dow down 405 points with the S&P 500 and Nasdaq lower, as Brent crude broke back above $1005 a barrel and the 10-year Treasury yield hit its highest level since 2023 despite a Treasury long-bond buyback of up to $6 billion6 that traders judged too small.
GPT-6 Astra scored 100%3 on ExploitBench and Siri AI now reads users' private messages and mail, yet the frontier model drawing safety scrutiny this week is Anthropic's Claude Mythos 5.1, reportedly released without pre-release evaluation by the UK AI Security Institute.
OpenAI's GPT-6 Astra, now rolling out across ChatGPT paid tiers, the API, Azure, and Bedrock, posts 98%1 on FrontierMath Tier 4, 99.9%2 on ARC-AGI-3, and 100%3 on ExploitBench. On Terminal-Bench Science 0.1 it leads Anthropic's Claude Fable 5.1 by 64.6%7 to 52.6%8 at roughly 31%4 lower estimated API cost, pressuring Anthropic on both capability and price.
The largest declared exchange of shipping strikes since the war began pushed Brent back above $1005 a barrel and sent US stocks down for a third straight session, with the Dow off 405 points. The 10-year yield climbed to a 2023 high as traders dismissed a Treasury buyback of up to $6 billion6 as too small to stabilize the market, a combination that sits uneasily with a Greed reading in market mood.
Apple Rents Siri's Brain for $1 Billion
Read full analysisA free Excel-formula helper that claims 340,000 users has rebranded itself as an agentic chat that finishes whole projects, and the public record for the new version is one Product Hunt review.
The spreadsheet muscle is real and the price is zero, but the agentic layer on top is still mostly a claim on a landing page. Everything users have actually praised in public is Excel formulas, Google Sheets scripts and translation, so the safe way to adopt it is as a free formula and data-cleaning assistant, with the report and app generation treated as a bonus to test rather than a promise to plan around. Finance and ops people living inside Google Workspace will get the most out of it; anyone outside Excel and Sheets should expect friction.
Solid AI for writing Excel formulas and for translating into other languages; the automation is useful and translation covers 28 languages.Paraphrased · G2
Ajelix is an agentic chat that takes a plain instruction and executes it end to end, producing standalone apps, polished reports, or automated workflows that connect tools and teams without code. It ships with a governance layer, audit trails, and guardrails, which is the part finance and operations leads should test first: give it one recurring report, then inspect every step it logged before trusting it with anything customer-facing. It has free options, and its Product Hunt rating is a 5.0 from a single review, so treat it as unproven until your own pilot says otherwise.
Try itSoftr AI generates a web application from one prompt and positions the result explicitly as a starting point to build from, which is a more honest framing than most prompt-to-app tools use. Use it to get past the blank-canvas stage on internal tools, client portals, or data-entry front ends, then spend your time on the permissions, data model, and edge cases the generator will not think about. Write the prompt as a spec with named user roles and the three screens that matter most; vague prompts produce generic apps you will throw away.
Try itRobin AI is an automated sales platform that uses AI and machine learning to handle prospecting and early outreach, the repetitive top-of-funnel work that eats rep hours. Before connecting it, write down your qualification criteria and the exact point where a human takes over, because an automated funnel with no handoff rule fills your CRM with noise. Run it on one segment for a month and measure reply rate and meetings booked against your manual baseline, not against the vendor's claims.
Try itMeta acquired Swedish startup Stilla.ai on September 9 to build out its Business Agent, which automates customer replies in WhatsApp, Messenger, and Instagram DMs and is already used by more than one million businesses since its June launch. Mark Zuckerberg has said the goal is agents that help run a whole business, so if your support or sales runs through Meta's messaging apps, decide today which questions the agent may answer alone and which must escalate to a person. Document those rules before the next capability update arrives, not after a bad automated reply lands in a customer's inbox.
Try it[EVAL/TRANSPARENCY] Anthropic's regulation case rests on internal evals , apply the 86979 provenance standard to Glasswing and 'When AI Builds Itself'87086.Description. KeilerHirsch The connection.
github.comReceipts come from the Firecrawl Developer Index (READMEs, docs, issues, merged PRs), first-party sources only. The paper comes from the Firecrawl Research Index, arXiv, last 7 days, picked for today's lead. Quotes are verbatim.
Hand Ajelix a bounded deliverable. Give it one instruction that covers a whole small project, then read the audit trail line by line before you accept the output. The trail is the product, not the answer.
Treat Softr AI output as scaffolding. Generate a working web app from a single prompt, then immediately list what you would refuse to ship: auth, data handling, and error states. Fix those by hand.
Define a qualified lead before switching on Robin AI. Write the disqualification rules first, in one page, then let it work the top of the funnel and sample ten conversations a day for drift.
Astra saturating FrontierMath Tier 4 and ExploitBench at roughly 31%4 below a rival's API cost tells you raw capability is no longer the moat. Apple's Siri strategy reportedly resting on Google's Gemini tells you where the real leverage sits: whoever owns the model layer under someone else's product owns part of that product. Anthropic's Mythos 5.1 skipping UK pre-release evaluation is the same story from the governance side, where speed is outrunning scrutiny. Build so you can swap the model underneath you, and keep an audit trail of everything the agent touched.
Method names the computation path behind the row, from a closed vocabulary: as published, yfinance close, 03B search corroboration, 04R dossier scoring, not recorded.
A figure is money, a percentage, a spelled magnitude, a multiple, a rate carrying its unit, a score over its denominator, a thousands separated number or an index close. Bare integers and years are not counted, so a phrase like "over 100 integrations" carries no mark. A chart's own scale line is the axis of the level above it and is counted once, with that level.
167 editions so far, one a day. Unsubscribe anytime.