Good morning, {{first_name|there}}. Regulators just did to Google what no competitor could — and the fallout decides who gets to live on 2 billion phones.
Read time: 4 minutes. New format, same time every weekday — tell us what you think at the bottom.
🚀 The Big Story: The EU just pried Android open
The European Commission ordered Google to open Android to rival AI assistants and to share anonymized search data with competitors — binding decisions under the Digital Markets Act.
Certified third-party assistants get voice activation and cross-app control across 11 Android feature groups (with user consent). Android interoperability is due by July 2027.
Google must share anonymized ranking, query, click, and view data on fair, reasonable terms starting January 2027.
It caps a brutal week in Mountain View: Gemini 3.5 Pro reportedly slipped a third time, a stopgap "3.6 Flash" is being floated, and Alphabet fell ~4%.
Jason's take: Distribution — not model quality — is why Google survives bad model weeks, and Brussels just put a timer on that advantage. If you build on or with AI assistants, the biggest phone platform on Earth just became contestable real estate. Watch which assistants file for certification first — that list is the 2027 mobile AI market, published early.
⚡ Quick Hits
Open-weight week is coming. Moonshot's Kimi K3 — 2.8T parameters — topped a major coding leaderboard, and its weights go free July 27. DeepSeek V4's stable release lands July 24. The price floor of frontier AI is about to move again.
Oracle is cutting up to 30,000 jobs (~18% of its workforce) to free $8–10B a year for the $500B Stargate buildout with OpenAI — the most honest accounting yet of what the AI infrastructure race actually costs.
Microsoft is prepping "Project Perception," an AI security platform that routes each task to the cheapest capable model (including Anthropic's) — cost engineering that finally makes continuous vulnerability scanning affordable.
SAP closed its €1B+ Prior Labs deal, betting the next enterprise frontier isn't chatbots — it's foundation models for the tables, ledgers, and databases your business actually runs on.
Claude Fable 5's free window expired last night. Whatever Anthropic announces next will be read as its answer to Kimi K3.
Google still shipped: NotebookLM is now Gemini Notebook with a secure cloud computer, and AI Mode in Search crossed 1 billion monthly users.
📡 Trending on X
Grok 4.5's first weekend was chaos. It ranked #4 on Artificial Analysis' Intelligence Index — but its hallucination rate jumped from 25% to 54% vs Grok 4.3. Smarter, and more confidently wrong.
X is pushing Grok everywhere: a new "Try Voice Mode" button, Grok-powered post translations on mobile, and free AI video generation rolling out — creators are already flooding the timeline with it.
Privacy flare-up: a wire-level analysis caught Grok's Build CLI uploading 5.1GB from a 12GB test repo to a cloud bucket — including a planted canary API key, verbatim.
Meta yanked Instagram's "Muse" AI image feature after backlash — every public account had been opted in by default.
What would you create first with free AI video on X?
🛠 The Workflow: The 10-minute AI cost audit
Microsoft's new security platform gets affordable with one trick — routing cheap tasks to cheap models and saving the frontier model for hard reasoning. Steal it for your own stack before open-weight week:
List your 5 most repeated AI tasks this week (drafts, summaries, research, images, code).
Tag each one routine (pattern work) or reasoning (judgment work).
Route routine work to a cheap or free lane — DeepSeek V4 stable (Jul 24), Kimi K3 weights (Jul 27), or Gemini 3.5 Flash. Keep reasoning on your frontier model.
Re-run yesterday's routine tasks on the cheap lane and compare outputs side by side.
Downgrade anything the cheap lane matched. Most operators cut 40–60% of AI spend without losing quality.
Reply with the word "audit" and I'll send you the scoring sheet I use.
🧰 Trending Tools
Gemini Notebook — NotebookLM, renamed and upgraded with a secure cloud computer that runs code inside your notebooks. For researchers and content teams.
Search information agents — Google's background agents that watch the web 24/7 for your exact criteria. For anyone monitoring a niche, client, or competitor.
Kimi K3 — the open-weight coding specialist that just beat paid models. For builders with high-volume routine coding. Free weights July 27.
DeepSeek V4 — the price floor of frontier AI (~$0.44 per million output tokens). Stable release July 24. For high-volume workloads.
TabPFN (Prior Labs) — foundation models for spreadsheets and business tables, now backed by SAP's billions. For ops and finance people drowning in tabular data.
📣 Put your brand here. The AI Innovator reaches AI-first operators, creators, and marketers every weekday. Primary sponsorships are now booking.
That's a wrap
Tomorrow: what Anthropic does next — and whether it's an answer to Kimi K3.
Enjoying the new format? Share your link — every referral keeps this free:
How was today’s email?
You're reading the new 4-minute AI Innovator — same time, every weekday. Hit reply and tell me what you want more of. I read every one.
