🔈 Listen on Spotify

Hi! Welcome to the 71st 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}

/REA /Coding /Skills

Ever looked at an app you pay for and thought, “Surely I could build my own version of this?” A developer known as Morluto has built a toolkit that lets an AI agent investigate the app and start rebuilding the bits you actually want. It is called Reverse Engineer Anything, or REA, and it gives coding agents the tools to look under the bonnet.

Reverse engineering is exactly what it sounds like: start with the finished thing and work backwards. Instead of describing an app from memory, you let the agent inspect it, work out what happens when you click a button or change an input, then help build your own version of that behaviour. A bit like taking apart a watch to understand the mechanism before making one yourself.

The documented examples are wonderfully specific. REA helped recover the acceleration rule behind a browser version of Chrome’s dinosaur game, then reproduce it in a mini-game with an adjustable speed. It also helped recreate the percentage-button rules in Windows Calculator. In the old brick-breaking game DX-Ball, it recovered the calculation that moves sound between your left and right speakers. These are working pieces of software, not evidence that every complete app can be copied perfectly.

According to the Perplexity announcement shared in our source material, REA is also now a skill in Computer, with a Figma-style editor among its examples. The instruction can start as simply as “Please reverse engineer Figma for me”. The interesting shift is being able to point at an existing product and say what you want to understand, rather than knowing how to code it first.

It brings some old questions back into focus. If your agent can inspect a logged-in app, what private information can it see, and where does that information go? And being able to reconstruct a feature does not settle who has permission to reuse its code, images or branding. Start with a small, authorised target and dummy data, not your client’s live account.

/Ramp /Agents /Productivity

You finally get AI to produce a proposal the way your business actually does it. Then your colleague opens a new chat and starts the whole explanation again. Your best instructions are sitting in your chat history, helping exactly one person. Sound familiar?

Ramp is the company that helps businesses manage corporate cards, expenses and supplier payments. On 13 October, it launched Glass, an AI workspace built around a simple idea: when one employee teaches AI something useful, the rest of the company should benefit too.

Glass brings together information from tools such as Notion, Slack and Google, alongside reusable skills: saved instructions for doing a particular job. Colleagues can join the same session rather than forwarding screenshots and starting again. Think of it as giving the team a shared working memory, with AI attached.

Ramp’s own example makes this easier to picture. An account manager saved a prompt that pulls feature requests out of customer calls. Another prompt prioritises them; a third emails customers when the requested feature ships. Ramp says it now captures about 20,000 requests a month, up from roughly 300 when people copied them manually.

And because AI bills can get silly, Glass also chooses which model handles each request. That is model routing: sending the work to the appropriate AI rather than reaching for the same one every time. A meter shows each project’s spending, with limits you can set.

Our take? This is a much more useful question than “which chatbot should everybody buy?”: how does one person’s good work become the whole team’s starting point? Glass is in early access, and South African availability is not specified. Check access, permissions and client-data handling before connecting the company’s filing cabinet.

/Google /Gemini /Reasoning

Every business has that folder. A client brief, three versions of a spreadsheet, meeting notes, a PDF someone forgot to name properly. Turning it into something useful takes longer than writing the final document. That is the sort of work Google is aiming at with Gemini 4 Argon, released on 13 October.

Argon is Google’s new top-end model for difficult jobs that involve several steps: reading the material, working out what matters, using tools and producing an answer you can actually do something with. Google says it is better at understanding documents that mix text and visuals, following writing instructions and sticking with complex coding work.

One of Google’s demonstrations is refreshingly close to a normal working day. In the Gemini app, Argon takes an untidy folder of project information and turns it into a client proposal deck using a saved skill. That is a more interesting test than asking it to write another generic email: can it pull the right facts together and turn them into something coherent?

There is evidence beyond the launch video. On the Vals Index, which tests finance, coding, legal and tax tasks, Argon ranked first as of 16 October. That makes it worth paying attention to. It does not tell you whether it will understand your particular spreadsheet, and a good ranking certainly does not make its answers safe to send unchecked.

You can try it in the Gemini app on Google AI Pro or Ultra; developers can use the paid API. The API introductory price is $2 per million input tokens and $10 per million output tokens until 30 November, then $4 and $20. Those are developer usage prices, not a new consumer subscription.

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

A welcome video without take seventeen? HeyGen’s Digital Twin turns footage of you into a reusable presenter.

Don’t believe us? Try it yourself…

Pro tip: Check pronunciation and lip sync. Label it AI-generated.

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

On 16 October, Anthropic published two reports on work at the AI frontier. Together, they ask an uncomfortable question: what happens to hiring when employees can hand more of the actual doing to agents?

Its employment study found that junior roles fell from 27% to 9% of technical hires. That is a change in the mix of hiring, not a claim that 18% of all jobs disappeared. Anthropic also tripled its total headcount over the past year.

The hiring signals are changing too. Mentions of candidates’ AI tooling experience tripled in technical hiring assessments and quadrupled in non-technical assessments between 2024 and 2026. Technical interviews placed more emphasis on autonomy and judgement, with less on narrow domain expertise.

The work itself increasingly involves directing agents, checking results and building the systems around them. Human attention becomes the constraint. Anthropic points to sharing productivity gains with employees and helping people move into new internal roles as important responses. Neither solves the missing first rung for juniors.

Then comes the expectations gap. Staff expect AI to be able to handle an average of 75% of their current work by 2030, compared with 28% among US workers. These are survey expectations, not a forecast that those jobs will vanish. People working closest to the tools may see further ahead, but they are hardly a neutral sample.

Our take? For a job seeker, “good at AI” needs receipts. Bring a small automation you built, explain the problem, show how you checked the output and describe where you still step in. A polished demo without those answers tells an employer very little.

For founders, make that a practical hiring exercise rather than another buzzword on the job advert. Give candidates a realistic task, let them use AI, and ask them to defend their decisions. Keep a route for juniors to learn the underlying work too. If nobody gets to practise judgement, where will your next experienced hire come from?

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 👋