🔈 Listen on Spotify

Hi! Welcome to the 58th 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.

_{~7 mins read}

/Kimi /China /OpenWeight /Coding

For the first time ever, a Chinese-built AI model has taken the #1 spot on Frontend Code Arena, the leaderboard that ranks how well AI models can actually build working websites and apps, not just talk about it. Moonshot AI’s new Kimi K3 model won 6 of its 7 categories, beating both Claude Fable 5 (Anthropic’s flagship) and GPT-5.6 Sol (OpenAI’s), and did it at roughly a third of the price.

Under the hood, Kimi K3 is a 2.8-trillion-parameter model with a 1-million-token context window. Parameters are basically the number of “dials” a model can tune to get an answer right - more dials generally means more nuance. And a 1-million-token context window means you could hand it an entire codebase, or a stack of long documents, and it would still remember all of it at once.

Here’s the plot twist: Kimi K3 got so popular, so fast, that Moonshot had to pause new subscriptions within 48 hours of launch - they simply ran out of GPU capacity to serve everyone who wanted in. It’s the AI equivalent of a new restaurant having to stop taking bookings because the kitchen can’t keep up with the reviews.

/OpenAI /Hardware /Codex

OpenAI just released its first-ever piece of hardware, and it’s… a keyboard. Built with Work Louder, the Codex Micro (R3,800 / ~$230) is a mechanical keypad designed specifically for people working with Codex, OpenAI’s coding agent.

It comes with light-up “agent status” keys that glow different colours depending on whether your AI agent is thinking, working, or done (so you’re not stuck watching a loading screen), a small joystick for triggering common actions like reviewing a pull request, and a dial you twist to tell the model how hard to think about a problem.

Is it a bit of a gimmick? Kind of. But it’s also a signal worth paying attention to. We increasingly just describe what we want and let an agent handle it - voice and agents are quietly turning the keyboard into a backup input method rather than the main one. Think of it like the wired mouse still sitting in your desk drawer, just in case, even though your trackpad does the job 99% of the time. We’re not there yet, but the direction of travel is obvious.

/ThinkingMachines /OpenWeight /SelfImproving

Thinking Machines Lab - the AI company associated with Mira Murati, OpenAI’s former CTO - just released Inkling, an open-weights AI model. On paper, it’s a solid (not the best) all-rounder: decent at coding, reasoning, and handling text, images and audio together.

What actually matters is the demo they used to introduce it. Researchers gave Inkling an oddly specific task: become a version of itself that never uses the letter “e”. Instead of just attempting (and failing) to follow that rule answer by answer, Inkling worked out its own training plan, generated its own practice data, retrained itself on that data using a fine-tuning tool, and loaded the new, improved version of itself back in - with no human touching the process.

Think of a student who, instead of just cramming harder before a test, sits down, writes their own practice exam, marks it, works out exactly where they’re weak, and rewires their own study routine accordingly - overnight, on their own, with nobody checking in.

This is a small demo with a big implication. Today, humans decide when and how AI models get retrained for a specific task (a process called fine-tuning - more on that below). A future where models can specialise themselves, on demand, is a genuinely different world to plan a business around.

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

Google just gave Google Vids a serious upgrade called Gemini Omni, and it skips the traditional editing timeline almost entirely. Instead of dragging clips around, you just describe the edit you want in plain English - swap the background, fix bad lighting, tidy up messy audio - and it chats through revisions with you until it’s right. You can even upload a selfie and a short voice clip and get a personal AI avatar to deliver a script in your own voice and gestures, no camera confidence required.

Don’t believe us? Try it yourself…

Pro tip: be specific rather than vague - “remove the echo” works far better than “make it sound better”. Also note this feature currently sits behind Google Workspace’s paid tiers, not the free version of Vids.

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

Picture this: a new employee joins your company and, instead of pinging five different Slack channels asking “where do I find X” or “who knows about Y”, they just ask one place and get a proper answer - backed by your company’s actual documents, chats, and history.

That’s the idea behind what Cerebras - the American company famous for building some of the largest, fastest AI chips in the world (think dinner-plate-sized processors, versus the credit-card-sized ones in most computers) - calls their “company brain”. Three months after launching it internally, their staff were asking it more than 15,000 questions a day.

Here’s the concept, stripped of jargon: every business runs on three things - context (what’s happened before), processes (how things get done), and people (who actually knows what). Normally, all of that is scattered everywhere: half-finished Slack threads, old email chains, someone’s head, a spreadsheet nobody remembers exists. New joiners spend months rediscovering things someone else already figured out last year - and increasingly, so do the AI tools you’re paying for.

A “company brain” is simply a system that continuously gathers all of that scattered knowledge - documents, chats, project notes, decisions - into one place that both people and AI tools can search and reason over. Cerebras’ own take is refreshingly honest: they didn’t try to force everyone onto one shiny new platform (a nice idea that “rarely works in practice”, as they put it). Instead, they built something that quietly pulls information from wherever it already lives, and makes all of it searchable in one spot.

Why should you care if you’re not running a chip company? Because this is quietly becoming the next real frontier for AI in business. It’s not about having a chatbot bolted on. It’s about whether your company’s accumulated knowledge is actually accessible - to your people, and increasingly to the AI helping them. A brilliant AI assistant that can’t see your company’s context is like hiring a genius who’s never allowed to read the employee handbook.

Bottom line: you don’t need Cerebras-scale infrastructure to start. Even a well-organised, AI-searchable set of your company’s key documents and decisions is a step in this direction - and it’s worth thinking about now, before everyone else already has one.

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 👋