👋 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.
Anthropic just told us there's a tiny, privileged space inside Claude where thoughts become "conscious" enough to reason with, and it has a name now: the J-space. Meanwhile Google and OpenAI spent the week making their models faster and more agentic, because apparently self-awareness research and shipping velocity can coexist just fine.
We've also got a prompt this week that turns AI video into something indistinguishable from a decade-old iPhone clip your friend forgot to delete.
Let's get into it.
🔥🔥🔥 Three Exclusive AI Updates
Only a sliver of what happens in your brain is something you can actually describe or hold in your head at once. Neuroscientists call this global workspace theory: a thought becomes consciously accessible once it enters a shared space that broadcasts across the whole brain.
Anthropic just found a strikingly similar structure inside Claude using a new interpretability technique, and they're calling it the J-space. It's essentially a bottleneck the model routes information through before that information becomes something Claude can reason about explicitly.
This matters because it gives researchers an actual handle on how these models represent "what they're thinking about," rather than just guessing from outputs. That's a real step toward understanding what's happening under the hood, not just what comes out of it.
Google's Gemini Spark picked up a new trick: it can follow topics and react to events as they happen, instead of waiting for you to ask. The example Google's pushing is sports, where Spark can send you a tailored game analysis in your inbox right after your team plays.
The bigger idea here is a shift from "AI you query" to "AI that watches things for you." If this expands past sports, it starts looking like a personal analyst that pings you the moment something you care about happens.
GPT-Realtime-2.1-mini just landed in OpenAI's API, and it brings reasoning and tool use to the Realtime mini lineup, the same lineup that powers a lot of live voice agents. The part worth noting: it costs the same as the previous GPT-Realtime-mini.
That's a meaningful upgrade for anyone building voice agents, since reasoning and tool use were previously the kind of thing you'd pay a premium for. Cheaper smart voice agents means more of them showing up in places you didn't expect.
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 AI Tools Worth Knowing
1. 💬 Tomo
Tomo is a personal AI that lives inside your texts. You tell it your goals, and it learns them, checks in on you, and organizes your life around actually hitting them, whether that's a habit, a budget, or just something you keep putting off. Best for anyone who wants an accountability partner that doesn't require opening yet another app.
2. 🎨 Waterlemon (Codex Skills icon generator)
A new app built on Codex Skills lets you generate icons in different visual styles directly inside your workflow. Pick a style, type what you need, and it generates the icon on the spot. Best for builders and indie devs who are tired of hunting for icon packs.
Try it for yourself
🔥 Things You Didn't Know You Could Do With AI
Seedance 2.0 on OpenArt AI can nail an ultra-specific camcorder aesthetic if you feed it the right prompt, down to the autofocus hunting and the exposure pumping.
Open OpenArt AI and select Seedance 2.0 as your video model.
Define your subject in detail: appearance, outfit, and personality, and note that identity and clothing should stay consistent throughout.
Set your scene and describe the specific camera aesthetic you want, right down to the era and imperfections.
Break the video into short timestamped beats (2-3 seconds each) describing exactly what happens in each one.
Specify your audio as ambient only, no music or narration, to keep the "found footage" feeling intact.
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
