In partnership with
News of the day
1. Microsoft unveils MAI-Thinking-1 and MAI-Code-1-Flash LLMs, notable for their smaller size and licensed data training. MAI-Thinking-1 outperforms Sonnet 4.6. → Read more
2. Walmart restricts employee AI tool usage due to high costs, implementing token limits to control expenses and measure ROI. → Read more
3. Microsoft introduces Scout, an AI agent for Teams that automates tasks, manages schedules, and drafts responses, acting as an always-on coworker. → Read more
4. Meta's AI agent for WhatsApp Business is now globally available, offering customer support, product recommendations, and appointment booking. It's also coming to Instagram. → Read more
Our take
Hi Dotikers!
Microsoft dropped seven new in-house models at once, but two stand out. MAI-Thinking-1, a 35 billion parameter reasoning model, said to edge out Sonnet 4.6 in blind human evaluations. And MAI-Code-1-Flash, 5 billion parameters, built for GitHub Copilot and VS Code, in the same vein as Haiku but cheaper.
The instinct is to stop at the benchmark. Except Sonnet 4.6 came out on February 17. Matching that model today is basically announcing: yes, on raw performance, we have caught up to where you were three and a half months ago. In the meantime, Anthropic shipped Opus 4.7, then Opus 4.8. The real feat is holding that level with a model small enough to run on a decent laptop, and cheaper. Not the benchmark itself.
What actually matters is the line slipped into the announcement: clean data, licenses in order, zero distillation from competitors' models. Microsoft is the first major lab to make data provenance a central selling point rather than a footnote. For anyone working in finance, insurance or any regulated sector, that changes the conversation. A capable model that does not drag a legal cloud behind it is exactly what risk committees have been waiting for these past two years.
Still to be verified. "Appropriately licensed data" sounds great on a slide, it is harder to audit in real life. Everyone helped themselves at the open web buffet without ever asking for the bill, and Microsoft shows up claiming it brought its own lunch. We will believe it when we see it.
The contrast is delicious. Microsoft is selling trustworthy models, humans always in control, right as the Meta story we covered yesterday resurfaces: a support bot wired into Instagram account recovery, handing access to anyone who asked nicely. The flaw had actually been sitting open for months, long enough for high-profile accounts like Sephora's to get hijacked. Trust on a slide and trust in production are two different things. That is exactly the gap this launch will have to close.
Alex.
2026 State of AEO Report
A year ago, most marketers weren't thinking about AI search. Now it's one of the fastest moving channels in the industry and nobody has a playbook yet.
So we built one. We surveyed hundreds of marketers to find out how they're approaching answer engine optimization, where they're investing, what's actually working, and what isn't.
The result is the 2026 State of AEO Report. Real data. Real strategies. A clear picture of where AI search is headed and how to get ahead of it.
Meme of the day
