- Practicaly AI
- Posts
- π§ Your Inbox Has a Brain Now β And So Does Your Browser
π§ Your Inbox Has a Brain Now β And So Does Your Browser
Today in AI: Gmail finally gets smart, Perplexity moves into Slack, and Google just scared every Bitcoin holder on the planet.
π Hello hello,
A Google research paper just made crypto Twitter spiral. Your Gmail inbox is about to stop being chronological. And Perplexity quietly turned Slack into something that does actual work.
Let's get into it.
π¬ Quick note: Weβre building something to help teams actually get good at AI (not just use it). β Get early access here
π₯π₯π₯ Three Highly Curated AI Stories
Gmail's new AI Inbox replaces the entire concept of scrolling through messages. Instead of a chronological list, you get two AI-curated sections:
"Suggested to-dos" (bills, deadlines, things that need action), and
"Topics to catch up on" (deliveries, updates, low-urgency stuff).
You open Gmail, and the decisions about what matters are already made.

The broader suite β AI-generated thread summaries, a writing assistant that matches your tone, and smart replies that don't sound like a robot β is already live for all users. The full AI Inbox is rolling out in beta for AI Ultra subscribers, US-first in English.
Here's why this isn't just a feature update. Once people get used to AI triaging their communication β deciding what's urgent, what's noise, what deserves a response β they won't go back. And with 3 billion Gmail users, Google is setting the default for how the entire world interacts with email. The company that controls the filter controls the attention.
Rollout expands to more languages and regions later this year.
Perplexity's "Computer" β a digital worker, not a chatbot β now lives inside Slack. You DM it or drop it into a channel, and it handles multi-step work: pulling data from Notion, filing tickets in Linear, drafting briefings from your emails, running research across the web and your internal docs. All from a Slack message.
But the real story isn't the tool. It's how it spreads. When someone uses Computer in a shared Slack channel, everyone in that channel watches it work in real time. Perplexity's own finance team automated accounts receivable this way. Their sales team followed with proposal generation. Nobody mandated it from the top β people saw it working and started using it themselves.
That's the pattern to watch: AI adoption that spreads laterally through teams, not top-down through IT rollouts. The companies that figure this out will move faster than the ones still doing "AI strategy" decks.
To be fair, this currently requires both a Perplexity Enterprise subscription and a paid Slack plan β so it's out of reach for small teams right now. But the adoption model is the thing to file away.

Parts of a quantum computer in Yorktown Heights, New York.
Photographer: Angela Weiss/AFP/Getty Images
The computing power needed to break Bitcoin's encryption with a quantum computer is 20 times lower than anyone previously estimated. That's the finding from a Google Research paper published March 31st. We're now talking fewer than 500,000 physical qubits to crack it β in minutes.
Here's the detail that should make you sit up: the paper describes "on-spend attacks" β a quantum computer fast enough could intercept a Bitcoin transaction while it's still waiting to be confirmed, before the block is even written. That's not a hypothetical future hack. That's an active transaction risk the moment the hardware catches up.
The machine doesn't exist yet. Current quantum hardware is orders of magnitude away. But Google's own internal deadline for migrating to quantum-resistant security is 2029 β and they're telling the entire crypto industry to do the same. Ethereum already has a roadmap for this. Bitcoin's upgrade culture moves considerably slower.
The question isn't whether quantum computing threatens crypto. It's whether Bitcoin can upgrade fast enough to matter. We'll be tracking this.
π₯π₯ Two Things You Can Do With AI
You don't need to be a developer to build a 3D interactive website anymore. Google AI Studio β the free, browser-based tool for building with Gemini β now generates full working web experiences from a prompt.
But first, a quick beginner intro:
It's still early β complex interactions need refinement, and you'll hit limits on what a single prompt can produce. But as a starting point for prototyping ideas you'd never have built otherwise, it's worth 10 minutes of your time.
Hereβs the workflow:
Open Google AI Studio at aistudio.google.com
Take a screenshot of the visual experience you're trying to recreate (or just describe it)
Upload the screenshot and write your prompt β e.g., "Create an image showreel in a spatial 3D experience with webcam and hand tracking, zoom in and zoom out"
Let Gemini generate the full interactive HTML experience
Open it in your browser, point your webcam, and control it with your hands
Bonus: try this with your last vacation photos for a personal spatial photo album. Also, hereβs a quick demo.
Most people use about 20% of the tools they pay for. Claude can now tell you exactly which features you're leaving on the table β personalized to your actual work.
Hereβs the workflow:
Open Claude Settings and enable "Generate memory from chat history"
Have a few normal conversations so Claude can build context on your work
Then send this prompt: "Based on what you know about me and my work, what are the best Claude connectors, features, or skills I should be using that I'm not currently using?"
Read what Claude recommends β and actually try one this week.ere
Is this you? Your team is using AI. But theyβre not getting better results.
Weβre fixing that. Join the waitlist to find out how.
π₯ One Pro AI Tip Worth Knowing
Most people use Cowork like a chatbot. That's why it underwhelms them.
Claude Cowork gives Claude direct access to your local files β is one of the most powerful AI tools available right now. You point it at a folder, describe what you want done, and walk away. It reads your files, makes a plan, breaks it into subtasks, and executes. Real deliverables appear in your folder: formatted spreadsheets, presentations, organized file systems, research reports.
Here are the tips that make the biggest difference:
1. Create context files, not longer prompts. Put everything Cowork needs to know about you β your role, your writing style, your preferences, your current projects β into text files in your working folder.
2. Turn on Extended Thinking. Go to Settings, enable Extended Thinking, and select Opus 4.6. This is off by default, and it's the single most impactful setting change you can make.
3. Use Projects for recurring work. Cowork Projects (launched March 20) are persistent workspaces where instructions, files, memory, and scheduled tasks stay together.
4. Start with a dedicated test folder - Cowork has real read/write access to whatever folder you share.
5. Schedule the boring stuff -This is the feature that turns Cowork from "faster assistant" into "employee who shows up on time."
We wrote a full deep-dive on setting up Cowork properly β context files, connectors, plugins, workflows, and the limitations I've hit. If you're using Cowork (or thinking about it), this will save you the weeks of trial and error I went through:
π¬ Quick poll: Whatβs one AI tool youβve tried recently that actually stuck?
Did you learn something new? |
Until next time,
Team PracticalyAI
Reply