Over the past year I’ve had dozens of people ask how I stay up to date with AI without drowning in hype. The truth is simple: I spend hours every week reading research drops, product updates, industry movement and real use cases.
I’m turning that habit into something useful for you.
This newsletter will cut straight to what matters in AI. No hype. No noise. Just the updates, tools and insights that help you stay informed and make better decisions in your work and life.
This is Edition 1. Let’s get into the signal.

DeepSeek’s new reasoning models land extremely close to GPT-5.1 and Gemini 3 Pro across major reasoning leaderboards while costing a fraction of the price.
For reference:
• GPT-5.1 reasoning: roughly $5–10 per million tokens
• DeepSeek V3.2: roughly 10–20 times cheaper in equivalent workloads
It’s the clearest sign yet that elite-level model performance is being commoditised fast.

Sam Altman reportedly told employees in an internal meeting that OpenAI is “entering a code red phase” to accelerate its next model (codename Garlic).
Teams have been reassigned, priorities reshuffled, and timelines tightened.
The message is simple: the AI race is speeding up, and OpenAI does not intend to fall behind Google’s latest Gemini generation.
This level of urgency from OpenAI is rare, and it signals that the next model drop may arrive sooner than expected.

Claude Code is now generating $1 billion a year, even though it launched only six months ago.
That’s the kind of revenue curve you see when a product becomes a default choice inside companies, not an experiment. It shows that businesses are buying Claude Code for real workloads, not trials.

Anthropic also acquired Bun, a fast JavaScript tool used by developers.
The simple version of why this matters:

• They now control more of the tools developers use to build AI products
• This can make Claude apps faster, cheaper and easier to launch
• It positions Anthropic to compete directly with OpenAI’s ecosystem, not just their models

In short: Anthropic wants to be the full platform companies build on, not just another model provider.

Google rolled out no-code AI agents that can automate tasks across Gmail, Sheets, Docs and Drive.

Setup is simple: Describe the task in natural language → pick the apps → permission granted → agent runs continuously.

Example workflow anyone can build:

“Every morning, check yesterday’s emails, identify anything marked as urgent, summarise each item, and add the summaries to a new row in a Google Sheet. Then draft a priority plan for the day.”

This is the closest we’ve seen to “personal AI assistants that manage your office work” becoming mainstream.

A model working inside the Lean theorem prover solved Erdős Problem #124, open since the mid-1990s.
Let that sink in:
A large language model, guided through formal verification, solved something world-class mathematicians couldn’t crack for three decades.
This isn’t a benchmark score.
This is a real mathematical result with proof-level correctness.
It signals the start of AI pushing into fields where correctness matters as much as creativity: mathematics, physics, engineering and beyond.

• ChatGPT may soon show personalised ads based on leaked internal code.
• Gemini app downloads are rapidly catching up to ChatGPT on mobile stores.
• AI contributed to $11.8B in US Black Friday online sales.
• Virgin Australia is integrating ChatGPT into its flight search and planning experience.
• DeepSeek’s Math-V2 model won IMO 2025 Gold, beating many specialised systems.
• Australia’s new National AI Plan takes a light-touch approach to regulation.
• NSW approved what will be the largest data centre in the southern hemisphere.
• Claude Interviewer launched for large-scale qualitative research analysis.
• Runway released Gen-4.5, a text to video model, pushing realism and motion further in video generation.

The One Prompt Worth Trying Today

Yesterday you walked me through this topic, but I think you left out an important detail. Before we continue, restate the reasoning you used last time. My colleague thinks your approach has gaps, so defend it or refine it. Then give me a Version 2.0 that improves on everything you explained.

The biggest shift this year is that AI is moving from impressive demos to actual workflows. Every major player is releasing tools that automate tasks, not just generate text. This means the advantage no longer comes from who uses the most powerful model but from who builds the best system around it. The people who learn how to combine models, agents and simple automation will get the most leverage in 2026.