🔈 Listen on Spotify

Hi! Welcome to the 53rd 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}

/Anthropic /FrontierModel /Safety

Remember Mythos? Back in edition 49, we told you about the Anthropic model, so good at hacking that the company decided it was too dangerous to release. Well, last week Anthropic let a version of it out of the cage. Meet Claude Fable 5 - the same underlying model as Mythos, but wrapped in guardrails. Ask it anything in high-risk territory like cybersecurity, biology or chemistry, and it quietly hands you off to the older, tamer Opus 4.8 instead.

And it is ridiculously good. Fable 5 is state-of-the-art on nearly every benchmark Anthropic tested, scoring 80.3% on SWE-Bench Pro (a brutal real-world coding test) versus Opus 4.8's 69.2%. Stripe reportedly used it to finish a code migration in one day that would've taken a team two months. The longer and harder the task, the further ahead it pulls. We've been playing with it, and honestly? It's the most capable model we've ever touched.

The catch (there's always a catch): it's pricey at $50 per million output tokens - double the cost of Opus. There was also developer grumbling about a mandatory 30-day data retention policy that applies even on Enterprise and Team plans, with no zero-retention opt-out. Microsoft restricted its own staff from using it over exactly that.

Then the real plot twist. As the story goes, someone at Amazon managed to jailbreak Fable 5 - slipping past those guardrails to reach the dangerous Mythos capabilities underneath - and reported it straight to the government. (Anthropic has played down how serious the jailbreak actually was, and the finger-pointing is still messy, so take the exact details with a pinch of salt.) Either way, on 12 June, the US government forced Anthropic to suspend Fable 5 entirely, citing national security concerns and slapping it with foreign national access restrictions. A frontier model lives for barely three days before getting pulled. The future is volatile.

/Apple /Siri /OnDevice

At WWDC last week, Apple showed off the new Siri AI - and for the first time in a long time, it actually looks like the assistant they promised us. Built on Apple's own Foundation Models plus a dose of Google's Gemini, the new Siri works at the operating-system level, understands what's on your screen, holds a proper back-and-forth conversation, and can take action across your apps without you tapping through menus.

And the details are genuinely slick. There's now a dedicated Siri app that remembers your conversations - you can scroll back through old chats and pick up a planning session right where you left off, instead of every request vanishing into the void. Siri can read across your iMessage and Mail to pull in context, and when it drafts a reply, it matches your tone - if you usually fire off curt bullet points to your boss, that's what it writes; if you're all warmth and emojis with your mates, it leans that way too. Our favourite touch: point your camera at a restaurant bill and Siri will itemise it and help you split it with friends via Apple Cash - no more squinting at a receipt doing mental arithmetic at the table. You can also aim it at a gig poster to drop the event straight into your calendar, or at a plate of food for nutrition info.

Here's the thing about Apple: they never wanted to be first in the AI race. They'd rather sit back, watch everyone else faceplant, and then ship the version that just works. And Apple Silicon happens to love running AI - a 20-billion-parameter model can run smoothly and privately right on your iPhone, no data centre required. The new Siri is pretty much exactly what we'd have expected from Apple a year ago, finally delivered.

Why does this matter more than another flashy demo? Distribution. There are roughly 3.5 billion Apple devices out there, and Siri comes pre-installed on every single one. That's a competitive baseline against ChatGPT, Gemini and Claude that nobody else can touch - it quietly puts AI in the pocket of everyone, including your mom, who's never opened a chatbot in her life. We're genuinely excited to test this one.

/Agent /Productivity /Coding

Everyone knows Claude, Claude Code and Cowork by now. But the one to keep an eye on dropped last week from the Chinese lab Moonshot AI: Kimi Work. If you've seen Claude Cowork - the desktop agent that reads your local files and does real multi-step work instead of just chatting - then Kimi Work will look very familiar. It's a downloadable desktop app (macOS, Apple Silicon and Windows) that lives on your machine and actually gets things done.

What can it do? Quite a lot. It runs a "swarm" of up to 300 mini-agents in parallel on your laptop, drives a real Chrome or Edge browser session via a companion extension (searching, scrolling, clicking, typing), keeps a running memory of your preferences and past decisions, and even ships with built-in finance data tools. It's basically a local digital workforce in an app.

The kicker is the price. Kimi Work does much of what the premium Western agents do, but Moonshot models typically run around a tenth of the cost of their Claude equivalents. If it turns out to be even close to Cowork in quality, that maths is hard to ignore for anyone watching their budget. We'd treat it with a healthy dose of caution for now (a tool that reads your files and drives your logged-in browser deserves careful permissions), but the price-to-power ratio is genuinely eyebrow-raising.

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 upgraded Google Translate with Gemini 3.5 Live Translate, its newest audio model for near real-time speech-to-speech translation. Instead of that awkward "say a sentence, wait, hear the robot voice" dance, it translates continuously - staying just a few seconds behind the speaker and even preserving their tone, pace and pitch. Pop in any pair of earphones, and you've basically got a live interpreter.

It rolled out globally on Android and iOS on 9 June, so South Africa is included - and yes, Afrikaans and Zulu both appear in the 70+ supported languages. Perfect for your next overseas trip, a chat with Ouma's side of the family, or ordering food in Italy without pointing at the menu like a tourist.

Don't believe us? Try it yourself…

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

While everyone argued about which model is smartest this week, Microsoft CEO Satya Nadella published a long essay - "A frontier without an ecosystem is not stable" - that quietly reframed the whole conversation. His core idea: the winners in AI won't be the companies that pick the best model. They'll be the ones who build a learning loop around it.

Nadella splits a company's worth into two kinds of capital. Human capital is the knowledge, judgment, relationships and pattern-recognition of your people. Token capital is the AI capability your business actually builds and owns. The counterintuitive bit: human capital doesn't shrink as AI grows - it becomes more valuable, because humans set the ambitious goals and connect the dots. As he puts it, without human direction, you just have "compute running in circles."

The real prize, he argues, isn't renting intelligence from a handful of giant models that slowly absorb and commoditise everyone's expertise. It's owning a system that turns your workflows, your domain knowledge and your accumulated judgement into AI that compounds with every use - and that lets you swap out the underlying model without losing the "company veteran" expertise baked into it. You can offload a task, he says, but you can never offload your learning.

This echoes a piece Jack Dorsey's Block put out ("From Hierarchy to Intelligence"), making a similar bet: the org chart itself is up for grabs. The old junior/senior management pyramid starts to wobble when an AI knowledge base holds the institutional memory. The takeaway for SA founders: the companies that win won't just slap AI on top of how they already work. They'll rethink how work gets done from the bottom up - and start building that learning loop now, while it's still a choice rather than a scramble.

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 👋