Hi! Welcome to the 36th 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.

_{~11 mins read}

/OpenAI /Codex /DeveloperTools

OpenAI just dropped the Codex desktop app for macOS, and honestly, it's a brilliant strategy. Instead of forcing developers to live in the terminal, they've built a proper interface for managing multiple AI coding agents at once.

Here's what makes this different: Codex has been around since April 2025, but you needed to be comfortable with command-line tools to use it properly. Now there's a desktop app that lets you orchestrate multiple agents working in parallel, review their changes in separate threads, and switch between tasks without losing context. It's like having a command centre for your AI coding team.

The real power comes from the skills system. Think of skills as instruction bundles that teach Codex how to reliably do specific tasks - deploy to Vercel, implement Figma designs, manage Linear projects, generate images with GPT Image, or create professional documents. OpenAI includes a library of curated skills, and you can create your own.

The app also supports worktrees, meaning multiple agents can work on the same repo without conflicts. Each agent gets an isolated copy of your code, so you can explore different approaches simultaneously. As an agent works, you can check out changes locally or let it continue without touching your local git state.

Pairing this with GPT-5.2-Codex (which many developers consider the best coding model right now) is a proper combination. For a limited time, OpenAI's even made Codex available to ChatGPT Free and Go users, and they've doubled rate limits across all paid plans. The battle for developer hearts is heating up.

/OpenAI /Agents /OpenSource

Plot twist: OpenAI just acquired OpenClaw, and before you get angry about another indie project getting swallowed by Big AI, this one's actually a win for everyone.

OpenClaw is that scrappy browser automation tool that went viral last month - the lobster emoji that could control your browser and automate tasks autonomously. Creator Peter Steinberger built it as a playground project, gained over 100,000 GitHub stars, and suddenly had the entire tech world paying attention.

Now, Steinberger is joining OpenAI to work on personal agents for everyone, and OpenClaw is moving to a foundation where it'll remain completely open source. OpenAI is sponsoring the project and has made strong commitments to keep it independent and thriving.

Here's the strategic picture: Meta acquired Manus for agentic AI. OpenAI acquired OpenClaw. But Anthropic? They haven't acquired anyone. The likely reason? They're already leading in autonomous agents with Claude's computer use capabilities and don't need to buy their way in.

What this tells us: personal autonomous agents are the new battleground. Not just coding agents - we're talking about AI that can handle your daily digital tasks: book flights, respond to emails, manage calendars, fill forms, research products, coordinate schedules. The kind of agent that works in the background whilst you're doing something else.

When three of the biggest players make big moves in personal agents within months of each other, pay attention. 2026 is shaping up to be the year autonomous agents go mainstream.

/Kimi /Moonshot /China /AgentSwarm

Whilst US media obsesses over the latest OpenAI drama and Anthropic product launches, Chinese labs are quietly shipping models that rival Claude Opus 4.6 at literally one-tenth the price.

Meet Kimi K2.5 from Moonshot AI. This thing is a beast. It's achieving global state-of-the-art scores on agentic benchmarks: 50.2% on HLE full set, 74.9% on BrowseComp, 78.5% on MMMU Pro, and 76.8% on SWE-bench Verified. For context, these are the metrics that matter when you're evaluating how well AI can actually do complex work autonomously.

But the really impressive bit is Agent Swarm mode - up to 100 sub-agents working in parallel on different parts of a task, coordinating 1,500 tool calls, running 4.5× faster than single-agent setups. Think of it like having a full construction crew instead of one person trying to build an entire house alone.

Moonshot's also launched Kimi Claw (their version of OpenClaw integrated natively into kimi.com) and Kimi Code for production-grade coding. The entire ecosystem is live, the model weights are open source on Hugging Face, and the pricing is aggressively competitive.

Here's what nobody's talking about enough: Chinese AI companies like Moonshot and DeepSeek aren't knock-offs. They're legitimate competitors delivering comparable quality to US frontier labs at a fraction of the cost, with massive context windows and increasingly sophisticated agentic capabilities.

This is the Cold War of the 21st century, except instead of nuclear arsenals, it's who can build the smartest, fastest, most capable AI agents. The race is far closer than most people realise. If you're building something in 2026 and you're not at least evaluating Chinese models, you're leaving money on the table.

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

Let me ask you something: Why are you still typing?

No, seriously. The average person types 40-60 words per minute but speaks 150-250 words per minute. That's 3-5× faster. Modern AI-powered voice dictation is accurate enough (95%+ in quiet environments) that there's literally no reason to be manually typing everything anymore.

If you're still keyboard-bound in 2026, you're working slower than you need to. Here's how to fix that.

The Three Options:

1. Wisprflow - The Premium Cloud Option

2. SuperWhisper - The Privacy-First Mac Native

3. OpenWhispr - The Free Open Source Option

How to Get Started:

For Wisprflow:

For SuperWhisper:

For OpenWhispr:

Pro Tips:

The best part? You can actually try all three and decide. But seriously, if you're still manually typing everything in 2026 whilst your competitors are dictating at 3× speed, you're literally working slower than necessary.

Stop typing. Start talking. Thank us later.

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

You know Spotify - 713 million users, 100 million tracks, the app you probably have open right now. What you might not know is that behind the scenes, they have a massive codebase problem.

Think about it: thousands of repositories, constant upgrades, framework changes, dependency updates, configuration tweaks across a sprawling software ecosystem. It's like trying to repaint a skyscraper that's still being built whilst people are working inside it. And with AI tooling adoption accelerating, that codebase is growing faster than ever.

For years, Spotify used something called Fleet Management to automate code changes. But there was a ceiling. Engineers had to write deterministic scripts using abstract syntax trees and regex patterns - basically, you needed specialised expertise to automate anything remotely complex. Simple changes? Automated. Hard, nuanced migrations that required understanding context and making judgment calls? Still manual, labour-intensive work.

Then they integrated Claude Agent SDK in July 2025 as a background coding agent, and everything changed.

Here's what makes this wild: Spotify now has an agent that operates autonomously from natural language prompts to merged pull requests. An engineer can literally type "Convert all Java AutoValue classes to Records across the entire codebase" in plain English, and the agent goes to work.

No specialised scripting knowledge needed. No regex gymnastics. No AST manipulation. Just describe what you want in normal human language.

The agent doesn't just make the changes - it runs formatting, linting, builds, and tests to verify everything works before submitting a pull request. It's like having an incredibly diligent junior developer who never gets tired, never complains, produces shippable code, and works 24/7 in the background.

The results?

Max Charas, Senior Staff Engineer at Spotify, put it perfectly: "Our engineers are now able to execute fleet-wide migrations at a pace that simply wasn't possible before."

The really clever bit is integration. Spotify didn't replace their existing workflows - they enhanced them. Engineers version-control their prompts in Git. An internal orchestration system triggers Claude Code to execute transformations across repositories. For one-off tasks, there's a Slack bot. For major migrations, it's part of the CI pipeline.

And they're not stopping there. They're tackling tech standardisation efforts that would be too complex otherwise - like enforcing explicit context propagation for all Java gRPC services throughout the company. A breaking change that would normally take hours of deep gRPC knowledge per service? Now Claude automates most of the implementation and the engineers just review.

Niklas Gustavsson, Chief Architect at Spotify, said Claude "has consistently delivered the strongest performance for large-scale code transformation work," which is why they've adopted Sonnet 4.5 as their default. They collaborated closely with Anthropic's Applied AI team throughout the process.

The lesson for the rest of us?

You don't need to be Spotify-sized to use this playbook. The principle is simple: identify the repetitive technical work that requires too much specialised knowledge to script but not enough judgment to need a senior engineer, then let AI handle it.

Code migrations. Documentation updates. Dependency upgrades. Test generation. Framework modernisation. These aren't exciting tasks - they're necessary maintenance that drains engineering hours and delays actual product work.

If Spotify can save 90% of its time on complex migrations and merge 650 AI-generated PRs monthly, what could your team automate? The robots aren't taking programming jobs - they're taking the boring, repetitive parts so humans can focus on the interesting problems.

And honestly? That's exactly how it should be. Nobody got into software engineering to spend weeks manually updating dependency versions across thousands of files. Let the robots handle the grunt work.

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 👋