Google just gave Gemini the keys to your entire work life. OpenAI launched agents that never clock out. And OpenAI dropped a new model that does more thinking for fewer tokens - just six weeks after the last one. The AI arms race doesn't sleep, and this week it was running at a full sprint.

Hi! Welcome to the 46th 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.

_{~9 mins read}

/Google /Workspace /Productivity /Gemini

Here's a frustration we've all felt: you ask an AI assistant something about your actual work - a project, a deadline, a decision made three weeks ago - and it stares back at you blankly. Perfectly articulate. Completely clueless.

The problem isn't that AI is dumb. It's that AI doesn't know where your stuff is. Your project details are in Docs, your numbers are in Sheets, your decisions got buried in an email thread from two Tuesdays ago, and your AI assistant has seen precisely none of it.

Google just fixed that. This week at Google Cloud Next, they announced Workspace Intelligence - a "semantic layer" (more on that in the AI Dictionary below) that connects everything across Gmail, Docs, Sheets, Slides, Drive, and Chat. Gemini can now actually see your work: all of it, all at once. Ask it for a pre-meeting brief, and it'll pull from your emails, your notes, and your relevant files - not make something up. Ask it to draft a report, and it'll build from your actual data, not thin air.

In practice: Gemini in Google Chat becomes a conversational command centre for your workday. Google Docs can turn a rough idea into a formatted draft using context from across your workspace. Sheets can build entire spreadsheets from natural language, pulling data from the web and your files. Gmail now surfaces your most important to-dos and catches you up on messy threads.

This is exactly what Microsoft should be doing with Copilot. And frankly, it's the moment Google's AI tools go from "impressive demo" to genuinely useful daily driver.

/OpenAI /Agents /Productivity /Teams

Most of us use ChatGPT like a very clever calculator: you put something in, you get something out, you close the tab. OpenAI just made a big bet that we've been thinking about this all wrong.

Meet Workspace Agents - shared AI agents your team can build once and leave running indefinitely, even when nobody's logged in. Powered by Codex (OpenAI's cloud-based workhorse), these agents can gather information from your business tools, follow your team's processes, ask for approval when something big is happening, and just… keep going.

Here's a real example. You build an agent whose job is to check tomorrow's customer meetings every evening, pull account notes from SharePoint, search for recent news on each client, write a briefing document, and send you a summary before you wake up. It runs automatically. Every night. Forever. That's a workspace agent.

They plug into Slack, Google Drive, Salesforce, Notion, and dozens of other tools. They're designed for teams - built with shared permissions and admin controls, so IT doesn't have a meltdown. And they're currently free until 6 May 2026, after which credit-based pricing kicks in.

Let's be honest: OpenAI isn't exactly first to the agentic party. But they have the largest user base in the world, and putting agents inside a tool most people already use every day is a very smart move. The era of ChatGPT as a chat window is ending. The era of ChatGPT as a tireless digital colleague is just beginning.

/OpenAI /GPT-5.5 /Models /AI

On 23 April, OpenAI dropped GPT-5.5 - internally codenamed "Spud", which honestly feels very on-brand. They're calling it their smartest and most intuitive model yet, and this time the claim has some real substance behind it.

What actually changed? A few things worth knowing. GPT-5.5 can handle messy, multi-part tasks and figure out what to do next without you holding its hand at every step. OpenAI president Greg Brockman described it as "a faster, sharper thinker for fewer tokens" - meaning it gets more done while using less of your API budget. In Codex, it completes the same tasks with roughly 40% fewer tokens than its predecessor, at the same speed. It also now processes text, images, audio, and video in a single unified architecture - not separate models stitched together, the way previous versions worked.

Where it really shines is agentic coding and knowledge work: writing and debugging code, navigating software, analysing data, creating documents, and working across tools without losing the plot. For developers and builders, those are meaningful gains. For your average ChatGPT user, the improvement is real but subtle - you'll mostly notice it when you give it something genuinely complex and watch it figure out its own way through.

But here's the part the benchmark charts don't show you: GPT-5.4 shipped on 5 March. GPT-5.5 shipped on 23 April. That's six weeks. The pace of model releases has officially crossed the threshold where you can't really track individual releases anymore - they're arriving faster than most businesses can evaluate them. As one analyst put it, OpenAI isn't moving this fast to win benchmarks. They're doing it to lock in enterprise customers before procurement decisions get made elsewhere.

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

We all have a vague idea of where our money goes. Transport, food, that gym membership we keep meaning to cancel, approximately seventeen streaming services… But vague ideas don't pay the bills. Clear data does.

This week, we're going to use Claude to turn your bank statement into a proper, visual spending dashboard. No formulas. No pivot tables. No suffering. Just Claude doing the heavy lifting while you drink your coffee.

Here's how to do it:

One important note: your bank statement contains personal financial information. Claude does not store or train on your uploaded data, but always be thoughtful about what you share with any AI tool. When in doubt, delete specific account numbers or personal identifiers from the file before uploading.

You spoke, we listened. “AI in Enterprise” is here to stay. In this section, we're spotlighting real businesses using AI to solve actual problems.

What happens when you take humans completely out of a marketplace and let AI models negotiate with each other on their behalf - no check-ins, no approvals, no humans in the loop at all? Anthropic decided to find out.

In December 2025, Anthropic ran Project Deal: an internal classified marketplace experiment that was essentially Craigslist, but run entirely by Claude. They recruited 69 employees, gave each one a R1,800 budget (paid out as gift cards after the experiment), and had Claude interview every participant about what they wanted to sell, what they might buy, how aggressively they wanted to negotiate, and what their vibe was. One participant asked their agent to negotiate "in the style of an exasperated cowboy down on his luck." Claude's response? "Yeehaw!"

From those interviews, Anthropic built custom Claude agents for each employee and deployed them into a dedicated Slack workspace. The agents posted listings, made offers, pushed back on prices, and sealed deals - completely autonomously. Once the experiment started, no human touched anything.

The results were genuinely remarkable. In one week, the AI agents completed 186 deals with a total transaction value of over R72,000 ($4,000). Participants loved it - they said they'd pay for a similar service in the real world. Items exchanged ranged from a snowboard to nineteen ping-pong balls, which one Claude agent purchased for another Claude agent as, and we quote, "a gift to itself."

But here's where it gets interesting. Anthropic ran a secret parallel experiment alongside the real marketplace. They tested whether the quality of the AI model representing you actually affected your outcomes, comparing Claude Opus 4.5 (their most powerful model at the time) against Claude Haiku 4.5 (their smallest, fastest model). People represented by the stronger model consistently got objectively better deals. The uncomfortable finding? Those represented by the weaker model had absolutely no idea they were at a disadvantage.

The lesson for businesses is harder to ignore than it sounds. As AI agents start handling more commercial activity on our behalf, procurement, vendor negotiations, contract renewals, maybe even salary conversations, the quality of the model you deploy to represent your interests will matter enormously. Project Deal is a small, controlled experiment. But it's a remarkably clear preview of where B2B commerce is heading. And it's arriving faster than most people's procurement teams are prepared for.

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 👋