{{current_date_full_with_abbr_day}}
3-minute read
3 AI UPDATES · 2 TOOLS · 1 WORKFLOW
New look, same PracticalyAI. We rebuilt the daily: shorter, cleaner, three minutes, same 3-2-1 underneath. Rate it at the bottom and reply if something's missing.
👋 Good morning. Three hikers on Mount Shasta asked Gemini what to pack for an eight-hour summit push. It underpacked them, and eight hours became two days, a descent in the dark and one wrecked knee. The sheriff's office called it a "critical misstep." The model has never been cold. Pack like it has.
IN 3 MINUTES, GET 1% BETTER THAN EVERYONE YOU KNOW:
ChatGPT learns your sign-off from your Slack. Free doesn't.
Gemini makes music beds now, for everyone.
Transcription just dropped to 10 cents an hour.
Garry Tan's Claude Code team, and three sites that fix generic AI design.
Turn a photo folder into an on-brand carousel.
3 AI updates
1. ✍ ChatGPT learns your sign-off
OpenAI is testing Writing Style in ChatGPT Work: connect Gmail, Google Drive, Slack and SharePoint, and it lifts your phrases, your sign-offs and your capitalization quirks into whatever you draft next. It lives under Settings, then Personalization, then Writing style. Two catches. It's going to a small group of users first, per BleepingComputer, and OpenAI hasn't said how long it keeps the samples it reads or whether they touch training. Free doesn't get it at all.
Do this: If you're on Work and it shows up, connect only the one app you actually write in, draft one real email, and read it aloud. If it sounds like you, keep it. If it sounds like you on a good day, definitely keep it.
Your "Best," is finally unemployed.
Would you let ChatGPT read your Gmail and Slack to learn how you write?
2. 🎵 Gemini ships a jingle machine
Google's Lyria 3.5 is live in the Gemini app for all users, globally, on web and mobile, plus the Gemini API, AI Studio, Flow Music and Vids. Pick or describe a genre, choose vocal or instrumental, choose short or long, and start from a template: backing track, brand jingle, ringtone. Google says the new model brings more expressive vocals and richer arrangements. No price was announced for the app, and no "Pro only" gate.
Do this: Next time you need a bed under a Reel, generate one in Gemini before you open the stock library. Time both.
Your next video bed does not need a royalty spreadsheet.
3. 🎙 Ten cents an hour to listen
Microsoft's MAI-Transcribe-2 is a speech-to-text model, not a chat feature. It handles 60 languages, tells speakers apart, stamps every word with a time, and outputs clean or verbatim text. Microsoft's own numbers: a 5.2% word-error rate on the FLEURS benchmark and up to 10× faster than OpenAI's GPT-Transcribe. The price is $0.10 per hour of audio, a limited offer through the end of the year, through Microsoft Foundry, the MAI Playground or OpenRouter.
Do this: Point one long call or webinar at it through OpenRouter and compare the bill to your usual transcription meter.
Your interview pile just got cheaper ears.
2 Tools worth knowing
1. 10 GitHub repos that run a startup like a team · free · open source
A roundup of ten open-source repos founders keep re-installing, each one replacing a job you'd otherwise hire for. The headliner is Garry Tan's gstack: 23 Claude Code tools that act as CEO, designer, engineering manager, release manager, docs and QA. The rest cover research, marketing skill packs and product staples. The catch: it's a shopping list, not a one-click seat. Install the two you'd use this week, not all ten.
2. 3 sites that fix generic AI designs · free · reference boards
Better prompts never fixed generic AI design. References did. Savee is Pinterest for designers, Typewolf shows real brand fonts so the model stops defaulting to Inter, and Variant lets you type an idea and scroll options before you generate. Show the model what good looks like instead of arguing with adjectives; our short walks all three. The catch: taste boards don't build the thing. They stop you building the wrong thing.
1 Workflow worth stealing
Build an IG carousel from a photo folder, inside ChatGPT
Built from a folder of photos. Eight slides, one prompt, then edits.
Part 249 of Things You Didn't Know You Can Do With AI. Three inputs and one plugin, and ChatGPT builds the carousel from your own photos:
Photos: put them all in one folder and give ChatGPT access to it (the Google Drive connector).
Topic and notes: tell it what the carousel is about and what you want in it: "You have access to some of my photos in that Google Drive folder. Based on that, can we create an IG carousel post like the highly engaging ones people create these days, on the topic: non-AI skills to learn in a post-AI world."
Brand identity: add your logo and colours, and ask for it branded but clean. “Make the carousel branded to PracticalyAI but keep the clean aesthetic."
Canva plugin: turn it on for more inspiration. That's what takes the design up a level, and the result comes back editable inside Canva.
The catch: the first pass is a draft. The reel's second prompt was, word for word, "I don't want a draft. I want the actual carousel." Full tutorial lands on YouTube this week, and every part of the series is in one spreadsheet at practicaly.ai/things.
Don't forget to rate today's post
That's your 1% for today. Reply with one word on the new look. The best complaint gets fixed.
Kushank, AI teammates, and the human team
PracticalyAI
