Sponsored by
Moda is the AI design agent with taste
Moda's viral launch hit 4.4 million views in two days. Tens of thousands of professionals signed up. Startups, agencies, forward-thinking brands and top firms are now using Moda to create brand-aligned slides, ad creative, reports, social carousels and more.
Most AI tools tend to create what we call "AI slop": repetitions of the same colors, layouts and fonts. And when you try to fix it, you get stuck in a loop of re-prompting.
Moda is different. Drop in your website URL, and Moda learns your brand from the ground up: your colors, your fonts, your visual language. Then it helps you generate pro-quality slides, docs, and marketing assets.
The best part? Every layer is fully editable on a real canvas, and exports to powerpoint, PDF and more.
Beginners in AI
Good morning and thank you for joining us again!
Welcome to this daily edition of Beginners in AI, where we explore the latest trends, tools, and news in the world of AI and the tech that surrounds it. Like all editions, this is human curated and edited, and published with the intention of making AI news and technology more accessible to everyone.
THE FRONT PAGE
OpenAI's New Jalapeño Chip Brings the Heat to Nvidia
TLDR: OpenAI just showed off its first computer chip, called Jalapeño, built with Broadcom to run ChatGPT faster and cheaper and to lean less on Nvidia.
The Story:
On Wednesday, OpenAI and Broadcom unveiled Jalapeño, OpenAI's first custom AI chip. It's built for one job: inference, which is the work a model does when it answers your prompts. OpenAI says it went from first design to a finished design ready for the factory in just nine months. That's quick. Most chips like this take a year and a half to two years. The team even used its own AI models to help with the design. Early lab tests, including with its GPT-5.3-Codex-Spark model, show the chip does more work while using less power than today's top chips, though OpenAI hasn't shared full numbers yet. Bloomberg reports it could cut the cost of running models by about half. The first chips should reach data centers late this year, with far bigger rollouts coming after, including giant gigawatt-size data centers built with Microsoft and other partners.
Its Significance:
Making your own chip used to be something only a few giants did. Google has its own AI chips. So do Amazon and Microsoft. Now OpenAI is joining them, and the reason is money. Renting Nvidia's chips is pricey, and OpenAI loses a lot of cash running ChatGPT for hundreds of millions of people. A cheaper chip it controls could help it spend less and serve more. This also isn't OpenAI's first reach into hardware. Last year it spent about $6.5 billion to buy io Products, the startup from famed Apple designer Jony Ive, to build a new kind of AI gadget. OpenAI doesn't just want to make the AI. It wants to make the chips, the products, and the devices that run it.
QUICK TAKES
The story: A leaker found hidden text inside Anthropic's Claude Code (version 2.1.190) that points to its most powerful model, Fable 5, returning with weekly use baked right into paid plans instead of costing extra. Fable 5 was pulled on June 12 after a U.S. government order over a possible jailbreak risk, but the signs now point toward a comeback.
Your takeaway: Hidden code isn't a promise, so don't mark your calendar yet. But it hints at where pricing is headed: instead of paying extra each time for a top model, you might get a set amount every week, like a phone data plan that resets.
The story: A proposed class action in California accuses BP, Circle K, 7-Eleven, Walmart, Marathon, and others of using an AI tool called Kalibrate to coordinate higher gas prices, which the suit says broke state antitrust law. The lawsuit claims the software pushed prices up by 6 to 30 cents a gallon across more than 1,700 stations.
Your takeaway: This is part of a bigger fight over "AI pricing." Landlord software RealPage was sued by the U.S. government in 2024 for allegedly helping set higher rents the same way. And last year, Delta Air Lines took heat from senators who feared it would use AI to charge each traveler a different price based on what that person might be willing to pay. Delta denied doing that. The shared worry: when AI sets prices, it can push them up, and you'd never know.
The story: Google added a "Select from screen" tool to Gemini in Chrome. You draw a box around any text or image on a page, and Gemini reads it and answers questions about it, no copy-paste needed. It's rolling out in Chrome 149 for signed-in users.
Your takeaway: Little by little, AI is becoming always on and always watching, sitting next to whatever you're doing and ready to jump in. Handy, sure. It also means more of what's on your screen gets read by an AI, so it's smart to know when it's looking and how to turn it off.
TOOLS ON OUR RADAR
🏠 HelpGenie Freemium: An intelligent customer support builder that allows you to create branded conversational agents that pull accurate answers directly from your existing help documents and knowledge bases.
📞 OnCallClerk Paid: An artificial intelligence receptionist that answers your business calls around the clock to book appointments and capture high quality leads without any human intervention.
🔍 Fixy Paid: A unique research tool that runs your queries across multiple artificial intelligence models at once to compare their answers and synthesize a single highly accurate final response.
📈 ProductBridge Freemium: A highly intelligent feedback agent that continuously monitors multiple platforms to collect and prioritize user suggestions based on revenue impact so your team can make smarter product decisions.
TRENDING
Your Hermes Agent Now Has a Pet That Does Nothing - Nous Research added 3,200+ animated mascots to its Hermes AI agent that react when it's thinking, working, or failing. They have zero effect on how the agent runs. It's basically a friendly Clippy you actually asked for.
Figma Brings Real Code Onto Its Design Canvas - At its Config 2026 event, Figma added "code layers," letting designers turn a design into working code with one click, plus built-in animation and AI tools. The goal is to erase the slow handoff between designers and engineers.
Google Builds 'Computer Use' Into Gemini 3.5 Flash - Developers can now build agents on Gemini 3.5 Flash that see a screen and click, type, and scroll across browsers, phones, and desktops. It's the companion to the Chrome update above, and it follows Anthropic's similar tool for Claude.
Perplexity Launches an AI Built for Lawyers - Perplexity's new Computer for Counsel handles legal grunt work like research, contract review, and citation checks, and links every answer back to its source. It joins Anthropic and Microsoft in chasing the legal market.
Anthropic Says Alibaba Copied Claude on a Massive Scale - Anthropic told U.S. senators that operators tied to Alibaba used about 25,000 fake accounts to quiz Claude 28.8 million times, then used the answers to train their own AI. The trick, called distillation, means teaching a weaker model by copying a stronger one's output.
MIT and Microsoft Make AI Agents Faster and Greener - Researchers built a system that designs and tunes AI agent workflows automatically, cutting the wasted computing, energy, and cost these multi-step systems usually burn. You just describe what you want in plain words.
TRY THIS PROMPT (copy and paste into Claude, ChatGPT, or Gemini)
Name any word or modern notion. Get it defined as it truly behaves: witty, cynical, and funny because it's true.
Build a single-file HTML app with vanilla HTML/CSS/JS. The Devil's Dictionary — give a word or modern concept, get a witty Ambrose-Bierce-style cynical-but-true definition. Persist entries to localStorage key 'devils_dictionary_v1' so the user builds their own growing dictionary.
Aesthetic: antique dictionary page. Cream paper (#efe7d4) with subtle texture gradients, dark brown ink (#2e2620), oxblood red (#8c2f23) accents, faded gold (#9a7d3c). Double-rule decorative page borders. Playfair Display serif for the masthead/headwords (drop-cap-red first letter), Spectral serif for body and italics. NOT dark-themed and no mono — small-caps serif labels instead, for an old-book feel.
Form: centered single word input with a Define button and a "surprise me" random link (pick from a built-in list of modern notions like networking, wellness, brunch, deadline, hustle).
System instructions: keeper of a Devil's Dictionary in Bierce's spirit — witty, cynical, satirical definitions funny because they hold an uncomfortable truth; sharp, literate, economical; skewer folly/vanity/institutions/modern absurdity but NEVER punch down at vulnerable groups or be cruel about real named people; humor from clever truth not meanness; no em/en dashes (use commas/colons). Return raw JSON: headword, pronunciation (mock respelling), pos (abbrev), definitions (1-2, primary is the star), usage (witty example sentence), epigram (optional satirical one-liner), epigram_attribution (invented absurd authority, never a real person), see_also (2-3 related words).
Render like a dictionary entry: headword + italic pronunciation in slashes + italic part of speech, numbered (or ·) definition blocks, a left-ruled "as used in a sentence" usage quote, a centered italic epigram with invented attribution, and clickable "See also" words that define-on-click. Below: "Your Dictionary" section listing all saved headwords alphabetically (dedupe by headword), each clickable to revisit. Copy entry button. Loading line "Consulting the lesser truths…".What this does: Type a word or modern concept (or hit "surprise me") and it returns a satirical dictionary entry in the spirit of Ambrose Bierce: a mock pronunciation, part of speech, a primary cynical-but-true definition that lands, an optional second sense, a witty usage sentence, and sometimes a short aphorism attributed to an invented authority. The humor comes from clever truth rather than cruelty, and it never punches down or mocks real people. Each "See also" word is clickable to define next, and every entry you generate gets saved into "Your Dictionary," an alphabetized lexicon that grows as you go. Styled like an aged dictionary page with oxblood ink. Saves to localStorage.
What this looks like:
WHERE WE STAND(based on today’s news)
✅ AI Can Now: Help design big parts of a real computer chip, which let OpenAI take Jalapeño from first design to factory-ready in nine months.
❌ Still Can't: Run multi-step agent tasks without wasting power and money, so MIT and Microsoft had to build a system just to make those workflows leaner.
✅ AI Can Now: See what's on your screen and take action, clicking and typing across browsers, phones, and desktops.
❌ Still Can't: Reliably tell a safe instruction from a hidden, harmful one planted on a webpage, which is why Google added prompt-injection defenses and asks for human sign-off before risky steps.
RECOMMENDED LISTENING/READING/WATCHING
Devs (2020) - TV Miniseries
Alex Garland's FX on Hulu miniseries about a software engineer (Sonoya Mizuno) whose boyfriend starts a new job at the secretive Devs division of a Silicon Valley quantum computing company run by Nick Offerman, and disappears on his first day. Garland wrote, directed, and shot all eight episodes himself, and the show looks and sounds like nothing else on television. A beautiful and patient meditation on AI, determinism, and free will.
Thank you for reading. We’re all beginners in something. With that in mind, your questions and feedback are always welcome and I read every single email!
-James
By the way, this is the link if you liked the content and want to share with a friend.
Some * designated product links may be affiliate or referral links. As an Amazon Associate, I earn from qualifying purchases. This helps support the newsletter at no extra cost to you and Amazon makes a tiny hair less.
Agents Don't Need Databases. They Need Scratchpads.
Agents spin up an idea, try it, throw it away.
ghost gives each task its own Postgres scratchpad. Fork in 58ms, isolated, delete when done.
No project limits. No shared state. 100 hours/month free.