🔈 Listen on Spotify

Hi! Welcome to the 73rd 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.

~6 mins read

/Claude /Voice /Productivity

On 29 October, Anthropic highlighted a useful iPhone update: you can speak to Claude through Apple CarPlay. Not another voice telling you to turn left. Think of a personal assistant in the passenger seat, helping you get your head around the working day before you arrive at the office.

The useful bit is what you connect before you leave. Claude’s voice mode can use your connected tools, just as a text conversation can. Connect your email and calendar, then ask: “What do I need to know before my first meeting? Has the client replied, and what did we promise them?” Rather than hearing a generic pep talk, you can get a briefing grounded in your actual work. Follow up out loud when something needs explaining.

And it needn’t stop at your inbox. Linear is supported, while the Microsoft 365 connection can bring in Outlook, Teams and SharePoint. That one needs a work or school account and administrator setup. The point isn’t to connect everything indiscriminately. Give Claude the information it needs to explain the project, the outstanding decisions and the awkward question you would otherwise remember halfway through the meeting.

The catch: voice is still beta, uses your normal allowance, and the Free plan permits one connected tool. Claude supports South Africa, but your car still needs compatible CarPlay. Set everything up while parked, keep the conversation simple and leave screen checks or approvals until you stop. Hands-free is not distraction-free. Our take? This is a much more useful commute companion than an inbox you cannot safely read, provided it helps you prepare rather than gives you another job at the wheel.

/Granola /Voice /Productivity

The borders between AI apps are getting rather untidy. Wispr Flow made its name turning speech into text, then added a meeting notetaker on 5 August. Granola came from the other direction: meeting notes first. On 29 October, it announced voice typing. Apparently, nobody is staying on their side of the office.

Here is how it works. Put your cursor in an email, a document or a Claude prompt, hold the fn key, speak and release. Granola turns the ramble into text, adding punctuation and paragraphs and removing the verbal furniture. The ums, the repetitions, the sentence you started three times. You are still writing the message; you have simply fired your fingers from the first draft.

The attraction is the price: unlimited words, free for all Granola users, including the Basic plan. That is not a verified promise of free forever. And typing is not doing: saying “send this to the team” produces those words, rather than secretly emailing your colleagues. For a founder with a long explanation in their head and a very short prompt on screen, that distinction is useful. Get the context down properly, then decide what happens next.

A quick honesty note: voice typing is a Mac beta. Readers asking to switch on iOS or wanting a keyboard got a stay tuned. Its reply about Windows and iOS says they are coming, and its product page also lists Android as forthcoming. It is not offline transcription either: audio and optional app context go to processing providers. Check the settings before dictating client information. Two apps competing for the same job is good news; two subscriptions you forgot to cancel is less exciting.

/Google /Gemini /Benchmark

Remember Gemini 4 Argon from edition 71? Google has already brought out the next name in the science cupboard. Gemini 4 Carbon, announced on 28 October, is aimed at difficult work that takes more than one clever answer: investigating a problem, using tools, trying another route and following the job through.

It arrives at number one on the Artificial Analysis Intelligence Index, scoring 59 on its updated version 5.0. That is an independent set of evaluations, not simply Google giving itself a gold star. But “top of this index” is not “best at everything”. Different tests measure different jobs, and even Google’s comparison table has rivals ahead on individual tasks. Choose your champion after seeing it do your work, not just win somebody else’s exam.

The interesting improvement is stamina with judgement. Google says Carbon is better at sustained coding and cybersecurity work, including finding and fixing weaknesses, and more likely to ask when instructions are unclear rather than invent an answer. Its announcement thread also highlights resistance to malicious instructions hidden in material the model reads. The reported zero successful attacks applies to one evaluated benchmark. It is not an invincibility cloak.

Now the sizeable catch: Carbon is initially going to selected cyber defenders through Google’s Fairwind programme. Wider developer, business and consumer access is still to come, with no general South African release date to circle. For most readers, there is nothing new to switch on yet. Our take? Better models are increasingly being judged on whether they can stay useful through a whole messy assignment, not just produce an impressive opening paragraph. That is progress worth watching, even while the rest of us wait outside the lab.

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: Clone your own voice in ElevenLabs

Turn a short welcome script into audio that sounds like you. ElevenLabs Instant Voice Cloning needs Starter or above: Starter costs $6 monthly, roughly R100 before tax.

Don’t believe us? Try it yourself…

1. Open Voices → + → Instant Voice Clone.

2. Record or upload one to two minutes of clear speech. Use your own voice, without music or background noise.

3. Confirm your rights and consent, then Save voice. Find it in My Voices, choose Use voice and enter a short welcome script.

Pro tip: Check pronunciation, label the result AI-generated and keep access private. A convincing voice is not proof of identity.

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).

The question needs a small correction. AI is changing jobs. What has not appeared is the simple version in which smarter models arrive and whole professions promptly disappear. A brilliant demonstration is not an employment statistic, however confidently somebody posts it.

Three signals in a16z’s 30 October charts complicate that story. First, news coverage of AI and job losses has cooled from its May peak. Second, explicitly AI-related layoff announcements in August and September were the lowest of 2026. Third, Indeed postings have rebounded: engineering by roughly 30%, software development by 26% and production/manufacturing by 13% from October 2025. The news and layoffs chart and hiring comparison are worth seeing together.

Now the honest read. Headlines measure attention, announcements measure what employers disclose, and postings are not hires. These are US indicators, not South African results, and software postings remain below their February 2020 baseline. A rebound is not a return to the old peak. Nor does it prove AI caused the increase. a16z is an AI investor with an optimistic argument to make, not the last word on employment.

Phil Chen’s The Next Frontiers of Autonomy, published on 29 October, offers a useful explanation. Agents can finish tasks without being ready to own jobs. He identifies three gaps: coordinating conflicting priorities across a company, learning the right lesson from unspoken feedback, and participating usefully in a live conversation. Sending your salesperson and finance manager faster assistants does not settle which customer gets credit terms. Somebody still has to make that judgement.

This is partly the old coordination problem Ronald Coase described, in new clothing. And capability is not permission: Anthropic’s controlled agent simulations show why handing over authority needs checks. Chen also points to video-labelling research and a radiologist’s perspective on the harder problem of learning physical work. Those are specific examples, not proof that human jobs are permanently protected.

Remember our edition 71 discussion of Anthropic’s hiring research? Its junior share of technical hires fell from 27% to 9%, even as the company grew. That is a change in hiring mix, not an economy-wide job-loss figure. It is still a warning that the first rung of the career ladder can become harder to reach before mass unemployment appears.

Our take? Don’t build a business plan around either “everyone is replaceable” or “nothing will change”. Pick one workflow, make the handovers explicit, train the people using AI and measure quality as well as speed. Keep someone accountable for the outcome, and give juniors room to learn. Better work for customers is a more useful target than a smaller team on a slide.

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?

✉️ Send us an email

💭 Drop us a message on LinkedIn

We’re building this with (and for) you. 🚀

See you next Tuesday 👋