👋 Hello hello,
❗Before diving in, this is for the business owners and entrepreneurs: we're building AI automation workflow guides: step-by-step, click-by-click, tool-by-tool so the repetitive work runs itself. Tell us what you'd automate.
OpenAI just built its first computer chip, spent nine months designing it, and named it Jalapeño. After the pepper. Somewhere a very serious engineering team had a very fun naming meeting.
Jokes aside, the chip is a real attempt to make ChatGPT cheaper and faster to run, and it wasn't the only big move. Google taught Gemini to read whatever is on your screen, and Microsoft's newest voice model finally learned to sound like it has feelings.
Let's get into it.
🔥🔥🔥 Three Exclusive AI Updates
1. OpenAI built its own chip (and named it after a pepper)
Until now, OpenAI has rented Nvidia's chips to run ChatGPT, like most of the industry. This week it unveiled its own, built with Broadcom and called Jalapeño. A chip is the hardware that does the actual thinking, and this one is built for a single job: inference, which is the moment a model answers your prompt, as opposed to the slow, expensive training that happens beforehand.
What does this mean for you? Every answer ChatGPT gives costs real money to generate, and reports suggest a purpose-built chip could cut those costs by roughly half. OpenAI even used its own models to help design the thing in nine months, blazing fast for an industry that usually takes years.
Large-scale rollout is planned for late 2026 with Microsoft and other partners. For you, that eventually shows up as snappier responses and cheaper AI tools.
Gemini in Chrome is Google's AI assistant built into the browser. Its new "Select from screen" tool lets you highlight any text or image in a tab and send it straight to Gemini, instead of typing out a description of what you're looking at. It lives in the "+" menu and works like a built-in screenshot grabber.
It's a small change that kills a real annoyance. You point at the thing you need help with, and Gemini already has the context. Google shipped it with Chrome 149, so you may need to restart your browser to see it.
The same day, Google gave its Gemini 3.5 Flash model "computer use" skills, letting AI agents see, reason, and click across browsers and apps on their own. The direction is clear: assistants that can watch what you're doing and act on it.
Text-to-speech turns written words into spoken audio, and for years the catch has been that it sounds flat and robotic. Microsoft's new MAI-Voice-2 is its most expressive attempt yet. It speaks 15 languages with real emotion control (excited, whispered, embarrassed, and more) and holds a steady voice identity across long stretches of audio.
The detail that stands out for us: it code-switches between Hindi and English (and Spanish and English) mid-sentence, the way bilingual people actually talk. Testers preferred it over the previous version 72% of the time.
Tone is the whole game with voice. A support bot that sounds cheerful while delivering bad news feels worse than plain text. Getting emotion right is what makes audiobooks, assistants, and accessibility tools listenable. It's live in Microsoft Foundry now, with a faster "Flash" version on the way.
Help shape Practically AI
Tell us what's working. Your feedback shapes what lands in your inbox, and it takes less than 2 minutes.
🔥🔥 Two Things You Didn't Know You Could Do With AI
This Youtuber built an AI travel influencer avatar of herself and "time travelled" to Tudor London and vlogged daily life in 1536, like a period drama crossed with a TikTok diary. It's a super fun look at where AI video generation is heading: full characters living inside fully built worlds. If you make content, this is your nudge to think bigger than the usual talking-head format.
One founder shared on X how he runs a daily Claude "skill" that reads his goals and to-do list each morning, DMs him a clean list to tweak, then prioritizes everything (P1 to P3, deadlines, deep work in the morning) and proposes a time-blocked schedule built around his real Google Calendar. He replies "book it," and Claude creates the events, color-coded, with no double-booking. If you struggle with focus or prioritizing, it's clarity on autopilot.
🔥 One Workflow Worth Trying
If you've used AI image or video tools, you know the frustration: your face looks slightly different in every single generation. HeyGen's new Look Packs fix that.
HeyGen builds a "Digital Twin," an avatar that looks and talks like you, and Look Packs (powered by its new image engine) generate a full set of polished looks in one tap that all stay recognizably you.
Here's how to try it:
Create your Digital Twin in HeyGen by recording a short video or uploading a few photos.
Open Look Packs and generate your first pack (it's free when you make your Digital Twin).
Choose the outfits and settings you want across the set.
Drop those consistent looks into your videos so every scene still looks like the same you.
Help shape Practically AI
Tell us what's working. Your feedback shapes what lands in your inbox, and it takes less than 2 minutes.
Don't forget to rate today's post
Until next time,
Team @PracticalyAI
