🔈 Listen on Spotify

Hi! Welcome to the 57th 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}

/OpenAI /Agents /Productivity

Remember Codex, OpenAI’s coding tool for developers? It just grew up, changed its name, and got a job. Last week OpenAI launched ChatGPT Work, a brand-new agent that lives on your computer (not just in your browser) and actually does things: it can read your files, create new ones, click around your apps, and stay on a single project for hours before handing you finished work.

The clever bit is where it runs. Because ChatGPT Work sits on your desktop rather than in a browser tab, it can reach into your actual files and apps and get its hands dirty. Think spreadsheets, slide decks, Word docs, or a full web app, built from one instruction while you go do something else. The old Codex app has been folded into a single ChatGPT desktop app, so Chat, Work, and Codex now live side by side.

If this sounds familiar, that’s because it is. This is OpenAI’s answer to Anthropic’s Claude Cowork and the wider race to turn chatbots into colleagues. OpenAI even announced a new model, GPT-5.6, on the same day to power it (more on that below). It’s available on desktop now for every plan, including the free tier, with web and mobile rolling out over the coming days.

/OpenAI /Models /Benchmarks

OpenAI also dropped a shiny new flagship model, GPT-5.6, and it’s a genuine step up. There’s a fun bit of backstory here: GPT-5.6 was first announced back in late June but kept under lock and key, with the US government limiting access to a handful of trusted partners for about two weeks. That restriction has now lifted, and the model is out in the wild for everyone.

OpenAI is shipping it as a family of three: Sol (the powerful flagship), Terra (the balanced middle child), and Luna (the fast, cheap one). The headline claim? On a tough test called Agents’ Last Exam, which measures how well a model handles long, real-world jobs on its own, OpenAI says Sol beat Claude Fable 5 by 13.1 points, and did it cheaper. On coding-speed benchmarks, it also comes out ahead.

Honestly though? It’s not the clean knockout the marketing suggests. Independent testers still put Claude Fable 5 ahead on broad, general intelligence and on complex, multi-file coding where deep reasoning matters. The fair read: these two are neck and neck at the frontier, and OpenAI led with the benchmarks it happened to win. What’s not in dispute is the price, with Sol costing roughly half of Fable 5 per token, which is exactly why the whole industry is paying attention. However, we must note that some people we trust a lot say that they feel that GPT 5.6 is superior to Fable in many regards.

/Voice /Dictation /Productivity

We’ll say it plainly: at TomorrowToday we reckon voice dictation is one of the biggest, most underrated AI unlocks going. Speaking at 150 words a minute instead of typing at 40 is a genuine superpower, and tools like Wispr Flow have proven how good it can get. The catch has always been the price, with the best ones tucked behind a monthly subscription of around $10 (roughly R180).

This week, Willow blew that gate wide open. They’ve released Frontier Mini, a free, unlimited dictation model that works system-wide across your Mac, Windows, or phone, in any text field, from Gmail to Slack to WhatsApp. It’s cloud-based with zero data retention, meaning your audio is processed and immediately binned rather than stored. In their own testing, they claim it’s faster and more accurate than WisprFlow, OpenAI, and Deepgram.

Is it quite as polished as the paid heavyweights for every edge case? The jury’s still out, and Willow does sell a pricier Frontier Pro model for power users. But if you’ve been dictation-curious and didn’t fancy paying to talk to your computer, this is a no-brainer. Free, unlimited, and genuinely good.

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

If you’ve ever built or designed anything with AI, you’ve probably noticed it looks... a bit off. Generic gradients, animations that feel clumsy, spacing that’s almost right but not quite. There’s a word for it now: slop. The good news is you can fix it in one move.

A designer called Emil Kowalski combed through Apple’s legendary WWDC design talks and distilled them into a single “skill”, a set of 17 design and motion principles that you can hand to your AI to instantly level up its taste. It covers the good stuff: how motion should feel physical, when animations help versus annoy, and the design foundations that make Apple’s interfaces feel like magic. You can use it to review existing work or build something new.

Don’t believe us? Try it yourself…

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

One of the most interesting stories in AI wasn’t a new tool. It was an argument. Alex Karp, the famously eccentric CEO of data giant Palantir, went on a proper rant (very on-brand for Karp) about the way most businesses use AI. Some people watched it and saw pure theatrics. We think it might be remembered as a pivotal moment, the point where the conversation started shifting towards something that’s only going to matter more from here: AI sovereignty. Strip away the jargon, and it asks one simple question: who actually owns your AI, and does it matter? The further we go down this road, the more we think the answer is yes, it matters a lot.

The argument goes like this. When your business runs on someone else’s model, you’re not just renting the software. You’re feeding it your data, your workflows, and your hard-won know-how, the stuff Karp calls your “alpha”. Over time, that edge quietly migrates from your business to the provider’s. Palantir laid this out in a nine-point manifesto, and two lines cut through: “Controlling your weights is controlling your fate” (your model weights being the distilled form of everything the AI has learned from you, see this week’s Dictionary), and a warning against “tokenmaxxing”, the trap of measuring progress by how much AI you use rather than what you actually get back, which they nicely describe as “the addictive feeling of false progress.”

Now, worth knowing: Palantir is not a neutral referee here. It has just launched a sovereign AI system with Nvidia that lets customers run models on their own hardware, so “own your AI” is also, conveniently, the thing Palantir is selling. A pitch dressed as a principle, if you’re being cynical. But a message can be self-serving and still be true, and the wider industry keeps proving the point: companies are running up enormous token bills for unclear returns, and governments from France to Spain are actively rethinking how much they want to depend on foreign AI providers.

So what does this mean for a founder or business owner? For most everyday work, renting AI from the big labs is completely sensible, and far cheaper than trying to build your own. You don’t need sovereignty to draft emails. But as AI moves closer to the core of what makes your business special, it’s worth pausing to ask who owns the model, the data, and ultimately the edge. Because at the end of the day, whoever owns the model owns the data.

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 👋