🔈 Listen on Spotify

Hi! Welcome to the 70th edition of the TomorrowToday newsletter.

We’re here to decode the AI chaos so you don't have to. Think of us as your friendly neighbourhood tech translators - we cut through the chaos, translate the jargon, and spotlight new AI tools that matter for founders, builders, and curious minds.

Buckle up, because the future's moving fast and we're here to make sure you don't get left behind! ⚡

If you enjoyed today’s newsletter, please forward it to a friend & subscribe by following this link.

_{~8 mins read}

/Grok /Agents /Automation

Remember the shared Grok bots from edition 69? Between 7 and 9 October, Grok Bot picked up three abilities that make an AI colleague feel more real: email, a view of what is happening on X, and a connection to your online shop.

First, an email address. Your bot can collect newsletters, receive reports and contact a supplier from an agent address, without borrowing your personal inbox. Think of giving your assistant a work email so it can handle the back-and-forth while you get on with your day.

The update that might have slipped past you is X search and monitoring. Instead of opening the app and disappearing down a rabbit hole, ask your bot to follow your industry, watch a product launch or tell you what customers keep complaining about.

Here is where that gets interesting for a business. One suggested approach is to split the work between three bots. A community manager watches brand mentions and flags conversations worth joining, giving your team a useful morning briefing instead of another tab to keep refreshing.

A content writer can take those conversations and turn them into ideas. Have it look at your previous posts, find the topics people responded to and draft in your voice. You choose what goes out, but you have somewhere better to start than a blank page.

Then give an X analyst the people question: who is engaging with us, and who should we get to know? It could surface a customer, a creator or an industry voice your team has missed, with the original posts attached so you can judge.

And if you sell online, the Shopify connection lets your bot check orders, track stock and help update product listings. Imagine asking what needs attention before opening the shop, and getting the low-stock items and order picture in one reply.

/Claude /Data /Visualisation

You know the meeting. Somebody shares a dashboard, everyone looks at the lines, and then somebody asks: so what are we meant to do with this? Getting the numbers onto a screen was only half the job. Making them mean something is the part that usually takes longer.

Claude Dashboards and Claude Motion bring those two jobs closer together. Dashboards connects to data warehouses such as Snowflake, BigQuery and Databricks. Ask a question in ordinary language, and Claude writes the query and builds the dashboard. You can open the query behind a chart and see when it was last refreshed, rather than simply trusting the picture.

Motion is the other half. Give it text, charts or images, and it turns them into an editable animation you can download as an MP4. In edition 68, we had you coaxing a motion-graphics showreel out of Claude. This makes that idea more useful for ordinary business communication: take your checked monthly figures and ask for a short explainer of what changed and why it matters.

Our take? The interesting bit is the move from showing data to helping people understand it. A sales manager could send a 30-second explanation before a meeting, leaving the meeting itself for decisions. You still need to check the maths and the story, of course. Animation does not make a wrong conclusion right.

The catch: both are in beta. Dashboards is on Pro, Max, Team and Enterprise; Motion currently needs Team or Enterprise. They use your plan's allowance, and Enterprise admins need to enable them.

/Google /Privacy /Productivity

Meeting note-takers have become one of the easiest AI tools to recommend. You can pay attention to the conversation instead of trying to type every useful sentence. But there is a question we do not ask often enough: where does all that conversation go afterwards?

For many tools, the answer involves somebody else's servers, sometimes in another country. That is worth thinking about when the discussion includes a client, a sensitive project or information you would not happily forward to a stranger. Google AI Edge Foresight takes a different route: the AI work happens on your Mac.

Here is the clever part. You jot down a few rough bullets while you listen, and Foresight uses the meeting transcript to fill them out into proper notes. It can also look through documents you connect and surface relevant information during the conversation. The aim is to help you stay in the meeting, rather than spend it hunting through folders.

Google's 6 October announcement says EmbeddingGemma 2 finds the relevant material and Gemma 4 helps write the response. Those are the models doing the work locally, with no cloud subscription costs. Google says personal files, audio and transcripts are not sent to a cloud AI service.

A quick honesty note: this is an experimental app optimised for Apple Silicon Macs. We would try it on an ordinary internal meeting before trusting it with something important. Local processing is a useful privacy improvement, but people should still know they are being transcribed, and the notes still need a human read.

In this section, we aim to spotlight an incredible AI tool or use case and guide you on how you can try it.

A watch is easy enough to use. Explaining how a wound spring keeps the hands moving at the right speed is another matter. You can read a page about gears and escapements and still have no picture of how they work together.

That is what makes Anatomy, a skill from Ryan, such a fun idea. Describe a machine and Claude Code builds an interactive explainer: something you can rotate, inspect and slow down while you work out what each part does. A watch is a good place to start, but think aircraft engines, turbines or any mechanism you have always wanted to understand.

The quickest way in is to play with the example on the site. If you want to build your own, you will need Claude Code and its usual plan or API allowance. The skill is open source; the AI doing the building is not unlimited free computing.

Don’t believe us? Try it yourself…

Pro tip: leave time to play and improve it. The author's watch took nearly two hours to build. The point is understanding the mechanism, so check its explanation as carefully as its appearance.

In this section, we're spotlighting real businesses using AI to solve actual problems.

Imagine your team builds a process around an AI model. It reads the incoming work, makes a recommendation and hands the next step to somebody else. It works well enough that people start depending on it. Then an update lands, and the same request gets a different answer, or a refusal. What happens to the business process you built around it?

That is the practical question underneath Satya Nadella's new essay, published on 10 October. He argues that companies should treat powerful models as insider risks. Give something access to important information and the ability to act, and you need controls around it, however impressive its answers are. A supplier's assurances do not make the consequences somebody else's problem.

The latest All-In discussion approaches the issue from another direction: Claude's constitution, model welfare and the debate about whether an AI could be conscious. Anthropic's constitution expresses uncertainty about that question. It also sets out the values and priorities used to shape Claude's behaviour, including circumstances where helping the user is not the highest priority. None of that proves consciousness. It does remind us that the model arrives with somebody else's decisions about how it should behave.

For a business, the philosophical argument can wait. The operational one cannot. A refusal does not require a model to wake up and develop a personality. Changes to its training, rules or surrounding software can be enough to alter a workflow your team relies on. Sometimes those changes improve safety. You still need to know what they mean for your business.

Nadella's answer is to put the controls outside the model: limit what it can access, keep records of what it does, test failures and make sure a person can stop it. Our take is to add a practical escape route. Keep your business knowledge and processes portable, and test important jobs with more than one model. Choose the tool that works best today, but avoid building a business that can only work if that one tool behaves exactly the same indefinitely.

AI is full of jargon, and we’re here to decode it. Each week, we’ll give you a plain-English definition of a buzzy term you’ve probably seen (but never fully understood).

Thanks for reading TomorrowToday! We’d love to hear from you:

➡️ What would you like us to cover next?
➡️ Have a tool or topic we should feature?

We’re building this with (and for) you. 🚀
See you next Tuesday 👋