Oh So AI • Edition 011 • Self-Improving AI
Edition Summary
Today: Google I/O opens with a new Gemini model and an Android rebuilt around it, xAI ships Grok Build to join the coding agent race, and OpenAI now lets ChatGPT connect straight to your bank accounts.
💬 Daily Quote
“The day’s noise always passes. The one hard thing you finished is what stays.”
📌 TL;DR
Today: Google I/O opens with a new Gemini model and an Android rebuilt around it, xAI ships Grok Build to join the coding agent race, and OpenAI now lets ChatGPT connect straight to your bank accounts.
🏆 LLM Leaderboard #AI #Models
The most popular and most capable models this week.
The top of the intelligence board has not moved in three weeks. GPT-5.5, Claude Opus 4.7 and Gemini 3.1 Pro still hold the same seats. The real movement this week is one layer up, in coding agents, not base models. xAI’s Grok Build just walked into a race that already had Claude Code and OpenAI’s Codex CLI in it. And DeepSeek’s new V4 line is climbing fast on the usage side, so expect the popularity column to shift soon. Most capable model is a settled question for now. Best agent is wide open.
Source: OpenRouter Rankings + Artificial Analysis Benchmarks
📰 Oh, So AI did this #AI #News
1. Google I/O 2026 starts today. A new Gemini model and an Android built around it.
Google’s biggest event of the year opens this morning, May 19, with the keynote at 10am PT. On the table: a major Gemini model update, an Android 17 rebuilt so Gemini sits at the centre instead of bolted on the side, a first public look at Android XR glasses, and Aluminium OS, a laptop platform that folds Android and ChromeOS into one. The first Googlebook laptops are expected from Acer, Asus, Dell, HP and Lenovo later this year. Per Android Authority.
If you design or build products, the assistant layer of the most used mobile OS on earth is about to change shape. Worth watching the keynote live before you plan your next app update. Read more at Android Authority #AI #News
2. xAI launches Grok Build. The coding agent race is now a three-way fight.
xAI shipped Grok Build this week, a coding agent that runs in your terminal. It writes implementation plans, edits files, runs shell commands and manages dependencies from plain English prompts, and it can spin up to 8 agents at once on a 2 million token context window. It scored 70.8% on SWE-Bench Verified, the standard test for this kind of work. For now it is in beta and limited to the $300-a-month SuperGrok Heavy tier. Per Engadget.
A developer can hand a messy bug to a terminal agent, read the plan it writes, approve it, and get a fix back without leaving the command line. Read more at Engadget #AI #News
3. Anthropic is closing in on a $900 billion valuation, ahead of OpenAI.
Anthropic is in talks to raise at least $30 billion at a valuation above $900 billion, per reports this week. If it closes, that puts the Claude maker ahead of OpenAI, valued at $852 billion in March. It is likely Anthropic’s last private round before a possible public listing, which could come as early as October. The money is going straight into computing capacity. Per CNBC and Bloomberg.
If your team builds on Claude, this round is what pays for the server capacity behind your API calls over the next year. Read more at CNBC #AI #News
4. OpenAI lets ChatGPT connect straight to your bank accounts.
May 15. OpenAI launched a personal finance feature in ChatGPT for Pro users in the US. Through a tie-up with Plaid, you can connect more than 12,000 banks and brokerages, Chase, Fidelity, Robinhood, Amex and more, then ask ChatGPT about your spending, subscriptions and upcoming payments. It shows a dashboard of your portfolio and money flow. The team behind it came from Hiro, a personal finance startup OpenAI bought in April. Per TechCrunch.
A founder can ask ChatGPT what they spent on software last quarter and get an answer pulled from real account data, not a guess. Read more at TechCrunch #AI #News
5. Apple will pay $250 million over its Apple Intelligence ads. You may be owed money.
Apple agreed to a $250 million settlement over claims it oversold Apple Intelligence and the upgraded Siri before those features actually shipped. If you bought an eligible iPhone in the US between June 2024 and March 2025, you can file a claim for $25 to $95 per device. A preliminary approval hearing is set for June 17. Apple admits no wrongdoing. Per TechCrunch.
Five minutes filling a claim form could put real money back in your pocket if you bought an iPhone 16 or 15 Pro on the AI promise. Read more at TechCrunch #AI #News
6. A new startup raised $650 million to build AI that improves itself.
Recursive Superintelligence came out of stealth this week with more than $650 million at a $4.65 billion valuation. GV and Greycroft led the round, with NVIDIA and AMD also in. The London and San Francisco team, fewer than 30 people, includes CEO Richard Socher, former chief scientist at Salesforce, and Tim Rocktäschel, an ex-DeepMind researcher. The bet: AI that studies its own performance and gets better with no human in the loop. A first system is targeted for mid-2026. Per The Next Web.
If this works, the tools you use could get better on a weekly cycle instead of waiting for a yearly model release. Read more at The Next Web #AI #News
Enjoying this breakdown?
Join 4,200+ others getting these simple translations every Tuesday and Friday.
🪄 Oh, So AI can do that?! #AI #Tools
1. Grok Build. xAI’s coding agent for your terminal.
Live in beta this week. Grok Build runs from the command line. You describe the job in plain English, it writes a plan, you approve or edit it, then it edits files, runs commands and handles dependencies. It can run up to 8 agents in parallel and holds a 2 million token context, so it keeps a large codebase in view. For now it needs the $300-a-month SuperGrok Heavy plan. Per eWeek.
A small dev team can run one agent fixing bugs, one writing tests and one updating docs at the same time, then review every plan before any code changes land. Read more at eWeek #AI #Tools
2. ChatGPT Personal Finance. Connect your accounts, ask about your money.
Live May 15 for ChatGPT Pro users in the US. Connect your bank, card and brokerage accounts through Plaid, and ChatGPT shows a dashboard of spending, subscriptions, upcoming payments and portfolio performance. You can ask plain questions, like where your money went last month. Intuit support is coming, which would let it answer tax questions too. Per 9to5Mac.
A solo operator can find every forgotten subscription draining the business account in one question, instead of digging through statements. Read more at 9to5Mac #AI #Tools
3. HubSpot AEO Sensor. See how AI search treats your brand.
Launched May 14, free, no login needed. AEO Sensor tracks how answer engines behave day to day. It shows a daily volatility score, weekly AI-referred traffic, and visibility and citation benchmarks by industry across ChatGPT, Gemini and Perplexity. Per HubSpot.
A founder can check on a Monday whether AI search sent more or fewer visitors to their category last week, and plan content around it. Read more at HubSpot #AI #Tools
⚡ Oh, So I can do this #SEO #WebDev
1. ChatGPT referral traffic just hit a 12-month low.
New data, flagged as HubSpot’s AEO Sensor went live: visits sent from ChatGPT out to other websites dropped to their lowest point in a year. People are getting their answer inside the chat and not clicking out. Per PPC Land.
If a chunk of your traffic came from ChatGPT links, expect it to keep shrinking. The fix is to be the source the answer is built from, not the link below it. Read more at PPC Land #SEO
2. Google has retired FAQ rich results. Time to clean your schema.
As of this month, Google no longer shows FAQ rich results in search for almost all sites. The FAQ markup many sites added years ago now does nothing for how you appear. Per Google Search Central.
Audit your templates. Old FAQPage schema is not harmful, but it is dead weight, and the page space you handed to FAQs for SEO can now go to content that still earns clicks. Read more at Google Search Central #SEO #WebDev
3. Remix 3 hits beta and drops React entirely.
Remix 3 is in beta preview and no longer needs React. It is now a full-stack framework on its own, bundling routing, sessions, auth, forms, uploads, server rendering and UI into one package, built directly on the web Fetch API. It is a preview, not production ready yet. Per Remix.
If you start a new project this quarter, there is a fresh full-stack option that is lighter and has fewer moving parts than a React meta-framework. Read more at Remix #WebDev
🎣 Top Hook Ideas
Hook 1: xAI just dropped Grok Build. That makes three terminal coding agents fighting for your keyboard: Claude Code, Codex CLI, Grok Build. The one you pick is becoming a real career decision. #Tools
Hook 2: A 30-person startup just raised $650 million to build AI that makes itself better, with no human in the loop. The race is no longer to build AI. It is to build the AI that builds AI. #AI
Hook 3: Apple is paying $250 million because it advertised AI features before they worked. If you write “AI” into your marketing, that number is the new price of overpromising. #AI
Hook 4: ChatGPT can now read your bank account. Spending, subscriptions, upcoming bills, all in one chat. The line between your assistant and your money just disappeared. #Tools
Hook 5: ChatGPT referral traffic just hit a 12-month low. People get the answer and never click out. Your job is no longer to rank. It is to be the answer. #SEO
Did 'Oh, So AI' help you understand something new today?
Forward this to a friend who is overwhelmed by AI.
Curated with ❤️ ByHarshal
Creative Director orchestrating AI workflows for founders' teams. Writing about productivity, design, and AI systems.
You made it to the end! 🎉
Subscribers got this days ago. Drop your email or join our WhatsApp community to get the next one first.