🔈 Listen on Spotify

Hi! Welcome to the 59th 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.

_{~5 mins read}

/Claude /Automation /Productivity

Anthropic has quietly shipped one of the most useful features we've seen all year. In the Claude desktop app (inside Cowork, its "AI colleague" workspace), there's a new option called Record a skill. You hit record, do a task on your screen exactly the way you always do, talk through what you're doing as you go, and Claude turns the whole thing into a reusable skill it can run again on demand. No prompts to write, no code, no fiddly setup.

Here's why that's a big deal. The hardest part of getting AI to help has always been explaining: writing out every step and every exception until you'd almost rather just do the job yourself. This flips it around. You show Claude once, and it learns not just the clicks but the reasoning behind them. Think pulling your weekly report, renaming a batch of files, doing your month-end filing, or formatting a spreadsheet the way your boss likes it. It's available now on the Pro, Max and Team plans, through the desktop app.

Credit where it's due: OpenAI got here first, with its Codex tool adding a similar "record and replay" feature back in mid-June. But Claude's version is beautifully simple and built for regular humans, not just developers, and we reckon this "show, don't tell" approach is the real unlock for anyone who's found skill-building a bit too techy until now.

/Google /Models /Efficiency

On 21 July, Google released two new Gemini models: 3.6 Flash and 3.5 Flash-Lite. Notice what's missing? The big, flashy flagship. Google's top-tier model (3.5 Pro) is still "cooking" in testing, so this drop is all about the workhorses, the models that quietly do the bulk of everyday AI work behind the scenes.

The headline here isn't a benchmark, it's the bill. The new 3.6 Flash gets the same job done using around 17% fewer tokens (the little chunks of text that AI charges you for), and Google dropped its price at the same time (output fell from $9 to $7.50 per million tokens. Flash-Lite is the tiny, lightning-fast one built to power the AI "agents" that churn through high-volume grunt work like sorting, searching and summarising.

The honest read: independent testers say these models are faster and cheaper, not necessarily smarter. And that's rather the theme of the whole week. The industry is quietly shifting away from "who has the biggest brain" towards "which brain is the right size for this job, at the right price". Boring? Maybe. But it's exactly what makes AI cheap enough to use everywhere.

/Claude /Models /Pricing

Late on Friday night, Anthropic dropped Claude Opus 5, and the pitch is refreshingly different. Instead of bragging about beating its rivals, Anthropic is beating itself: Opus 5 gets close to the intelligence of Fable 5 (its most powerful public model) for about half the cost. It's especially strong at coding and real "get it done" agentic work, the kind of tasks that actually finish a job rather than just start it.

For most people, the takeaway is simple. If you're on Claude Pro or Max, Opus 5 is now the strongest model you can use, and it's the new default on Max, at no extra cost. For the builders paying per token, it lands at the same price as the old Opus 4.8 ($5 and $25 per million input and output tokens), which is half of what Fable 5 costs. There's even a new "effort dial" that lets you trade a little quality for more speed and a smaller bill.

One honest caveat: Fable 5 is still the smartest Claude for the very hardest, longest jobs, and it keeps the frontier crown. Opus 5 isn't trying to be the genius you save for special occasions. It's trying to be the reliable one you leave switched on all day without watching the meter. In a week where everyone's talking about cost, that's a very clever place to aim.

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

We just told you Claude can learn a task by watching you do it, so let's actually try it. Pick one boring thing you repeat every single week and teach it to your AI once, so you never have to explain it again.

Don’t believe us? Try it yourself…

Pro tip: The narration is the magic. Your clicks tell the AI what you did, but your voice tells it why, and that's what lets it handle next week's slightly different version without breaking a sweat.

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

The biggest enterprise story this week wasn't a product launch. It was a line drawn in the sand. On Friday, a coalition of around 25 American tech heavyweights (Nvidia, Microsoft, Meta, IBM, Dell, Palantir, Hugging Face and Y Combinator among them) published a joint letter titled “Open Weights and American AI Leadership.” Nvidia boss Jensen Huang thought it important enough to break his silence and publish his first-ever post on X just to share it. Microsoft's Satya Nadella backed the same message the same day.

The argument, in plain English: the West won't keep its AI lead by guarding one giant, secret “frontier” model. It'll keep it by building an open ecosystem, where “open-weight” models (ones anyone can download, inspect, tweak and run on their own machines) spread AI into every factory, hospital, farm, classroom and corner shop. Openness, they argue, actually makes AI safer (more people can spot and fix the flaws), cheaper, and far less dependent on a handful of gatekeepers. Huang's one-liner: the world needs both frontier closed models and frontier open models.

Nadella added the angle we keep banging on about: don't reach for the biggest, priciest model for every single job. Use the right-sized model for the task, wrap it in good tools and context, and stay in control so you're never locked to one supplier. Microsoft is already quietly routing work to its own cheaper in-house models wherever they match the expensive frontier ones.

Now the honest caveat, because it matters. This isn't quite “all of big tech agrees.” The companies that sell the big closed models (OpenAI, Anthropic and Google) pointedly did not sign. And Nvidia sells the chips that all this AI runs on, so “open AI everywhere” happens to be very good for Nvidia's business too. A message can be self-serving and still be true.

Why it matters here at home: for South African founders and businesses, open models are the road to sovereignty and sane costs, running genuinely capable AI on your own terms, on your own data, without a Silicon Valley meter ticking in the background. It's the same “who actually owns your AI?” question we raised a couple of editions back, now playing out on the world stage.

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 👋