OpenAI just dropped their most capable model yet, Microsoft rebuilt Copilot using Claude's own technology, your computer can now work while you sleep, and a teenager just sold his calorie-counting app for hundreds of millions of rands. Buckle up.

Hi! Welcome to the 39th 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}

/OpenAI /GPT5 /ProfessionalWork /AIAgents

OpenAI shipped GPT-5.4 this week, and it's their biggest model update yet. Here's why this matters even if you never plan to read a benchmark.

Every few months, these models get meaningfully smarter. Not tweak-smarter. Genuinely, measurably better at doing real work. GPT-5.4 is specifically designed for professional tasks - spreadsheets, presentations, long documents, multi-step research - and it's doing that work at a level that's starting to match or beat junior professionals in head-to-head comparisons. OpenAI tested it against industry professionals across 44 different occupations, and the model won or tied in 83% of comparisons. That number was 71% just one model ago.

What does that mean practically? It means the model is now good enough that if you're not using it, someone who is has a genuine edge over you. Not because it's magic - but because it handles the time-consuming, cognitively expensive parts of knowledge work faster and more accurately than ever before.

A few things worth knowing about GPT-5.4 specifically: it has native computer-use capabilities (it can operate a desktop, click through websites, fill in forms), it supports up to 1 million tokens of context (meaning it can hold an entire book in its head while working), and it hallucinates significantly less than previous versions - individual claims are 33% less likely to be false compared to the previous model.

It's available in ChatGPT now for Plus, Team, and Pro users. If you're on any paid plan and you haven't tried it yet this week, open a tab and give it something real to do.

/Anthropic /ClaudeCode /Automation /Scheduling

Anthropic just launched scheduled tasks in Claude Code desktop - and if you haven't started using Claude Code yet, this is the feature that should push you over the line.

Here's the idea: instead of asking Claude to do something now, you schedule it to run automatically on a timer. Every hour. Every morning. Every Friday at 5pm. While your computer is on and you're doing something else entirely - or sleeping.

The use cases are genuinely exciting once you start thinking about them. You could set Claude Code to check your error logs every few hours and automatically open a fix request for anything actionable. You could have it pull your email every morning and prepare a structured brief of what needs your attention. You could have it monitor a deployment, babysit a slow-running build, or remind you to push a release branch at a specific time. You set it up once, in plain English (or using the /loop command), and it just runs.

The scheduling is flexible - you can say "every 20 minutes", "every weekday at 9am", "check back in 45 minutes" - and Claude translates that into a schedule automatically. Tasks live inside your current session, so they reset when you close the app, but for anything persistent, you can hook it into GitHub Actions or use the Desktop version's durable scheduling.

If you run any kind of repeating workflow - daily reports, weekly summaries, monitoring jobs, client check-ins - this is worth at least 30 minutes of your time to explore. You can get a lot of hours back.

/Microsoft /Copilot /Anthropic /M365 /AIAgents

Microsoft launched Copilot Cowork yesterday, and the name is not a coincidence.

Earlier this year, Anthropic released Claude Cowork - a product that lets you delegate complex, multi-step tasks to an AI agent that plans, executes, and delivers finished work on your behalf. It went viral. It triggered a R5 trillion selloff in enterprise software stocks as investors realised that an AI that does the work might replace many of the tools currently used to manage that work. Microsoft's own share price dropped nearly 9% in February.

Microsoft announced Copilot Cowork. Built in close collaboration with Anthropic, using the same agentic technology that powers Claude Cowork - Microsoft confirmed this explicitly, saying they "brought the technology that powers Claude Cowork into Microsoft 365 Copilot." Same underlying model. Same approach to multi-step task execution. The key difference: Copilot Cowork runs in the cloud inside Microsoft 365, grounded in what Microsoft calls Work IQ - a layer of intelligence drawn from your actual work: emails, meetings, Teams chats, files, and calendar. Claude Cowork runs locally on your device. For large organisations with enterprise data and compliance requirements, that cloud-first approach is a significant advantage.

Here's what it can do today: clean up your entire calendar and reschedule conflicts, build a full meeting packet including a briefing doc, slide deck, and draft follow-up email, research a company across web and internal sources and deliver an executive summary plus a structured Excel workbook, and produce a complete product launch plan - competitive analysis, value proposition doc, pitch deck - from a single instruction. All running in the background while you focus elsewhere.

The Microsoft-Anthropic partnership runs deep. Claude models are now available in mainline Copilot Chat, GitHub Copilot, Microsoft Foundry, Excel, and PowerPoint. Copilot Cowork is in limited research preview now and rolling out to the Frontier program later this month.

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

This week’s challenge: Start speaking to your computer - and watch the hours stack up

What if the single biggest productivity gain available to you right now wasn't another AI model or a new workflow? What if it was just… talking? We have recommended this tool before but felt it is worth emphasizing it in this section again because it is just that good.

WisprFlow is a voice dictation tool that works across your entire computer - Gmail, WhatsApp Web, Word, PowerPoint, Notion, Slack, anywhere you type. You press a key, you speak, and it transcribes accurately, auto-formats your text (full sentences, paragraphs, punctuation), and drops it exactly where your cursor is. That's it. No switching apps. No copying and pasting. Just speak.

The reason this matters more than it sounds: most people type significantly slower than they think. And more importantly, they edit as they type - second-guessing phrasing, backspacing, rewriting. Speaking skips all of that friction. You say what you mean, Wispr cleans it up, and you're done. Users regularly report doubling the number of emails they get through in a day - not because the emails are worse, but because the process is faster.

It also handles multiple languages, including Afrikaans, which makes it genuinely useful in a South African context where you might be drafting something in English and dropping in a phrase in Afrikaans without thinking about it.

TomorrowToday recommends: this is the one AI tool that will save you the most time this year. One hour to set up. Hours back every week.

Here's how to get started:

Pro tip: Don't speak in bullet points. Speak in full thoughts, like you're explaining something to a colleague out loud. Wispr handles the structure. The more naturally you talk, the better the output.

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

This week: The 18-year-old who built a calorie app and sold it for millions

This week's story isn't about a big company rolling out an AI pilot programme. It's about two teenagers who built something, let AI do the heavy lifting, and just walked away with a life-changing outcome.

Cal AI is a calorie tracking app with a simple premise: take a photo of your food, and it tells you what's in it. No manual logging. No searching through a database. Just point your camera at your plate and let the AI figure out the rest. The founders - Zach Yadegari and Henry Langmack - built it while still in high school. Within under two years, it had been downloaded more than 15 million times and was generating over R550 million in annual revenue.

The product works because of AI - the core feature (photo-based calorie estimation) would have been impossible to build at this speed and cost even five years ago. The models that power it are trained on food imagery at scale. The interface is deliberately minimal because AI handles the hard part. Two teenagers with a small team of seven employees built something that a large company's entire product division couldn't have shipped as quickly.

MyFitnessPal - the biggest name in nutrition tracking - noticed. They'd been watching Cal AI climb the App Store charts for over a year. The deal closed in December. Terms weren't disclosed, but with R550 million in revenue and that kind of growth trajectory, the outcome for the founders almost certainly runs into the hundreds of millions of rands. Yadegari is 19 and still running the app, now as a unit of MyFitnessPal, while attending college.

Here's the larger point: this is going to happen again and again. AI has dramatically lowered the cost of building. A small team with the right idea, the right tools, and the willingness to ship quickly can build something that competes with - and gets acquired by - industry giants. We believe AI is going to be one of the biggest wealth creation events of our generation. The Cal AI story is a preview of what's coming.

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 👋