SUBSCRIBE | ADVERTISE

August: AI news sans hype

A monthly digest of the 1% of AI news and tools that mattered

Was this email forwarded to you? Sign up here.

To help deal with the endless stream of AI news and releases happening, I’m switching to sending out a monthly news recap of the top links you should check out. My goal is to make it easy for you to stay up to date on what’s going on — on your terms — without hype and overwhelm.

You’ll still hear from me on a weekly basis; I’ll keep sharing exactly how I’m using AI from week to week, and show you what’s actually working.

TLDR; GPT-5 was OpenAI’s messiest launch yet, Microsoft follows Meta’s playbook with an aggressive hiring spree, Anthropic enters the browser wars and Google has the internet going bananas.

New models

• The turbulent release of GPT-5 was the most talked about AI news in August, partially because OpenAI withdrew all other models and implemented an unreliable model switcher that sometimes gave you dumb answers; lots of these things were later fixed (including bringing back GPT-4o).

• Anthropic launched Claude Opus 4.1 the first week of August—an improved version of their flagship model.

• Nano Banana (aka Gemini 2.5 Flash) is Google’s new native image generation model, and has had the internet going bananas this last week. You can use it for free in the Gemini app or Google AI Studio; for builders, it’s also available in the API. To understand Nano Banana’s full capabilities, check out this full walkthrough featuring Logan Kilpatrick from Google.

• OpenAI’s released a real-time speech-to-speech model in the API. It processes and generates audio through a single model and API, which gives lower latency and higher quality audio. The Realtime API can now also use MCPs —connections to third-party tools — and has image understanding. Worth a look if you’re building voice apps with AI (unless you’re Taco Bell’s drive-thru).

• OpenAI finally launched some open weight models: gpt-oss-20b and gpt-oss-120b. They’re designed to be run locally and on-device, and have an open-source software license. These models aren’t frontier-pushing in terms of performance, but they allow devs to build AI into apps that benefit from running locally, e.g. privacy-focused apps or where internet access is not possible or convenient. If the models are good enough for their use case they might be a great choice in terms of price, speed and privacy; e.g. simpler use cases for tool calling logic or enabling execution of defined workflows within an app.

IN PARTNERSHIP WITH DOOLA

Start and scale your business with doola’s business-in-a-box.

Doola helps entrepreneurs around the world form their LLC, manage their books, and file taxes—all in one place. No stress. No second-guessing. Trusted by thousands of founders across 175+ countries, doola handles the boring stuff so you can focus on what matters: building your business

Plus with their new AI-Cofounder feature, you can get back-office business and tax advice tailored for your specific needs.

Get 10% off with code WHATSBREWING or get a free consultation.

Industry moves

• Sam Altman had dinner with a bunch of media outlets. Reveals OpenAI has better models they’re not releasing, draws parallels between AI hype and the dotcom-bubble, and says they might bid on Google Chrome if it’s forced to be sold.

• Following Mark Zuckerberg’s talent raid earlier this summer, Microsoft went on their own recruiting offensive with multi-million dollar offers for a list of targets from notable AI labs—including Meta’s AI division. Meta has since been busy reorganising the division and announced a hiring freeze.

• OpenAI rolled out a low-cost subscription plan, ChatGPT Go, available only in India. It gives expanded access to ChatGPT’s most popular features. Might be released in more countries.

• Elon Musk’s company xAI sued OpenAI and Apple, claiming their partnership is an antitrust violation. Apple, on the other hand, is reportedly in talks with Google to have Gemini power a revamped version of Siri. When you think about it, it’s strange that Apple hasn’t developed their own model yet.

• Nvidia and AMD have obtained US Gov permission to resume selling AI processors to China. Chinese authorities, on the other hand — helped by funds and tax incentives — are urging companies to buy domestic GPUs.

• Perplexity is rolling out a revenue-sharing initiative for publishers, who will get paid when their articles get traffic via Perplexity's AI browser, Comet. This comes amidst a bunch of lawsuits on copyright violation from large media companies.

Research

• Anthropic released a report with common use cases of AI for educators, including the type of apps they build with Claude Artifacts.

• A new study shows less junior roles are hired for in firms that have hired for at least one AI project.

• The leap from GPT-4 to GPT-5 was as big as the one from GPT-3 to GPT-4, but it might not seem so because of all the interim releases from OpenAI before GPT-5.

New tools & product features

• a16z is back with an updated version the top 100 GenAI consumer apps. Vibe coding apps saw a big boost (Lovable, Cursor, Replit, Bolt).

• Anthropic launched a Chrome extension: Claude for Chrome. Currently only available to Max subscribers. I tried it, and wasn’t impressed. It was extremely slow and just kept going in a loop on the simplest of tasks (e.g. adding text to a Notion document). Will be interesting to see how it progresses and eventually gets out of Beta.

• Projects in ChatGPT now have project-only memory. ChatGPT will remember only what you have in the project, without mixing it up with all your other memories. It also won’t carry the project’s context into your regular conversations. It’s a meaningful upgrade for use cases like creating reports, doing research or managing client work in a contained space.

• Microsoft has introduced a COPILOT formula for Excel. Lets you do AI operations (e.g. “categorise this”) and results update automatically as data changes. Currently rolling out to Microsoft 365 Beta Channel users.

• OpenAI’s Codex-CLI now supports image uploads, has an IDE extension, and a few other new features. Image uploads is super useful when vibe coding for going from design/mockup to coded interface. The IDE extension lets you run Codex in the sidebar of your code editor, such as VS Code and (my personal favourite) Cursor.

  • For the uninitiated, Codex-CLI is a coding agent that you operate through the terminal, similar to Anthropic’s massively popular Claude Code. I tested Codex for a few days back in May, but Claude Code was my clear favourite and I’ve stuck with it since. Codex now runs on GPT-5, too, so it’s closing the gap to Claude Code in terms of performance and UX—I might give it a shot again soon.

What I’m actually using

Out of all these headlines, here’s the stuff I’m using on a daily basis and how:

• GPT-5 with thinking mode enabled, inside ChatGPT. Using this when I want max accuracy, it’s not very emotionally intelligent but great at analysis, math and careful planning.

• I’m using Claude Opus 4.1 inside Claude for writing, and inside Claude Code for planning out PRDs and tasks lists before I actually start coding. Really happy with it so far.

What’s on my radar

• Definitely going to check out Project-only memory for ChatGPT. I already have several projects in ChatGPT, but it’s not possible to change the memory setting, so I’ll apply this to my future Projects.

• I’ll also experiment with Claude for Chrome when it eventually gets better (and out of beta)

• I’ll test the updated Codex-CLI soon, would be great to have an alternative for Claude Code with GPT models that I can use with my ChatGPT subscription.

• I haven’t used image gen a lot in my work lately, else I’d be all over Nano Banana.

PS thanks to the readers who booked 1-on-1 sessions last week—you are amazing. Your challenges really span the spectrum and are shaping how I think about what to cover in this newsletter. If you're stuck on how to apply AI to your life or work, I have 2 spots left this week. Details here | Book your session.

THAT’S ALL FOR THIS WEEK

Affiliate disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes link to products and other newsletters. Please assume these are affiliate links. If you choose to subscribe to a newsletter or buy a product through any of my links then THANK YOU – it will make it possible for me to continue to do this.