Mon, October 12, 2026
3-minute read
3
AI UPDATES
2
THINGS YOU DIDN'T KNOW YOU CAN DO WITH AI
1
FROM PRACTICALYAI LABS
👋 Good morning.
Somebody asked an AI to read 328 posts about their app last week so they wouldn't have to. That was the launch demo for Grok Bot's X update, which now searches, reads and monitors X without you connecting an account. Hand off the scrolling. Keep your thumbs for the coffee.
IN 3 MINUTES, GET 1% BETTER THAN EVERYONE YOU KNOW:
Grok Bot reads X for you. No account linking.
OpenAI's fastest Sol yet, if your plan qualifies.
Midjourney's model now thinks before it draws.
Pose the shot first, then let AI render it.
The Claude skill that edited Kushank's video on Friday.
3 AI updates
Grok Bot searching X for app feedback. Source: @bot on X.
On Oct 7, the @bot account said Grok Bot can now search, read and monitor X for every user, no X connector required. Ask it to track feedback on your product, follow a breaking story or send you a weekly roundup of your industry. Your own timeline and mentions still need the connector, and there's no published limit on how far back it searches or how often, per TeslaNorth.
Do this: Point it at your product name, or a competitor's, and ask for a Monday-morning summary of complaints and feature requests. You need a Grok Bot plan (SuperGrok or an eligible Cursor plan), not an X developer setup.
Reading the replies, minus the reading the replies.
Your AI can now read X for you. What would you have it watch first?
OpenAI is rolling out Ultrafast mode for GPT-6.1 Sol in the API, Codex and ChatGPT Work: up to 8× the speed of Sol Standard with what OpenAI calls near-Astra intelligence. The API price is $12 per million input tokens and $60 per million output, which OpenAI puts at 1.2× Astra. In Codex and ChatGPT Work it's limited to Pro 500, eligible usage-based Enterprise and credit-based Edu plans, and Enterprise admins have to switch it on.
Do this: Pay for it only when waiting costs more than the tokens: a production bug, an agent stuck in a long checkout flow, a live feature where every second shows. For everyday drafting, Sol Standard is plenty.
Speed, like everything else in AI, now has a tier.
Midjourney is testing Thinking Mode on its alpha website. Open a finished job in the lightbox, click "Rerun (Thinking)" in the lower right, and the model re-plans the image before it generates again. Midjourney says this improves prompt accuracy, typography and coherence, and it's collecting examples of where it helps and where it doesn't. It's alpha-site only for now, and may later become a full thinking mode or an "add more thinking" option after a job.
Do this: Rerun the image that keeps misspelling your product name. Text inside images is one of the three things Midjourney says improves, so it's the fastest test of whether the rerun earns its wait.
Your poster's typo finally gets a second opinion.
2 things you didn't know you can do with AI
1. 🖌️ Pose the shot before the AI draws it · ArtCraft · free, open source · Mac, Windows, web
Instead of rerolling prompts until the hands come out right, ArtCraft lets you stage the image first: pose a character in 3D, cut the background, drop in a mesh, and the model renders what you blocked. It's a free, open-source studio built to mirror Adobe's apps, PhotoCraft for Photoshop and LightCraft for Lightroom among them, per Futurism, with the code on GitHub. The catch: much of it is early alpha from a four-person team, so keep client files and logins out of it until it earns some trust.
2. 📄 Make a spreadsheet write its own formulas · GenOffice · free, open source · Mac, Windows, Linux
Open a real .xlsx in GenOffice, describe the tracker you want, and it writes the formulas, charts and pivot tables for you. Same trick in the other editors: Slides builds a deck from a one-line topic, and the PDF editor lets you change text in place or convert the file to Word on your own machine. It's a free, open-source office suite from Genspark, and the AI runs on Genspark credits or your own API key. The catch: it's at version 0.11, so try it on a copy, not the quarterly board deck.
1 From PracticalyAI Labs
Friday's Labs session was Kushank taking a raw clip to a first cut with Claude, live: cleanup, captions, his own fonts and style, motion design. The workflow he ran, in three moves:
Set it up: Treat the AI like a new editor you're onboarding: a strong model running in an agent, a video skill that writes the edit as code and renders an MP4, and Descript to strip filler words and long pauses from the raw take.
Brief it: References, not templates: a mood board, your own fonts pulled from a screenshot, and a short brief you dictate out loud the way you'd brief a person.
Give notes: The first render is never right. Watch it, leave timestamped notes like "At 0:07 the caption covers my face," and re-render in the same thread.
Inside PracticalyAI Labs: the Claude skill that ran the cut, the captions and the motion design was shared live on Friday, and it lives in Labs next to the replay and the slide deck. Drop it into your own Claude setup and skip straight to the brief.
Your camera roll has been waiting for this longer than you have.
LAST FRIDAY IN LABS
Get the Claude skill that edited Kushank's video
On Friday, members watched Kushank take one raw clip to a first cut with Claude: cleanup, captions, custom fonts, his own style, motion design. The skill that did it is inside PracticalyAI Labs, next to the replay, the slide deck, a live build every Friday and help when you get stuck. Your first week is on us.
Been a member before? Rejoin here.
Don't forget to rate today's post
That's your 1% for today.
Kushank, AI teammates, and the human team
PracticalyAI
Running a small business and want to build AI automations? Book a call with my trusted delivery partner.
