Hello, AI Explorer!
Good Wednesday. This week: Demis Hassabis put a year on AGI -- 2030, give or take one -- and named four things still standing in the way. The FDA fast-tracked 10 AI-designed drug candidates into Phase I human trials, the first formal federal program to do so. And Dropbox founder Drew Houston stepped down as CEO after 19 years, saying there has never been a better time to start building something in AI.
Welcome back to your weekly AI digest: human-curated, zero fluff.
Every Wednesday and Saturday, we bring you:
- The week's most important AI developments (no fluff)
- Hand-picked AI tools actually worth your time
- Expert insights on what these changes mean for you
Let's dive into this week's discoveries! ⚡
🔥 This Week in AI
🔥 This Week in AI |
May 27, 2026 |
| 🧠 | Google DeepMind's CEO says AGI will arrive by 2030 -- give or take a year |
| 🔬 | FDA fast-tracks 10 AI-designed drug candidates into Phase I human trials |
| 📦 | Dropbox founder Drew Houston steps down after 19 years to build AI |
| ⚖️ | Stanford: AI hiring tools show "clear racial disparities" across 4 million applications |
| ⚡ | OpenRouter raises $113M at a $1.3B valuation -- now processing 25 trillion tokens per week |
| 🎵 | ElevenLabs ships Music v2 -- commercially licensed, multilingual, mid-track genre switching |
| 🚫 | Trump scraps AI executive order -- voluntary 90-day model review dropped over China concerns |
| 🌍 | China locks down travel for top AI researchers at Alibaba and DeepSeek |
| ⚡ | 4 AI Tools This Week: Windsurf, Freepik, Fireflies, Cleo -- all inside. |
GOOGLE DEEPMIND
| 🧠 |
Demis Hassabis Just Named AGI's Year. He Said 2030. |
Google DeepMind CEO Demis Hassabis sat down at Google I/O and said something most AI researchers still won't put in writing: AGI is coming by 2030, give or take a year. He has said versions of this before, but his tone at I/O was different. Where he used to hedge, he now speaks in years.
A few unsolved gaps remain -- world physics modeling, persistent memory across sessions, behavioral consistency, and continual learning. Drug discovery is his first target: oncology and immunology get solved first, then a general disease engine for everything else. After AGI, he says he plans to use it to study the nature of reality and what it means to be human.
What Hassabis actually said:
• AGI target: 2029-2031, with 2030 as the center estimate
• Unsolved gaps: world physics, persistent memory, consistency, continual learning
• First disease targets after AGI: cancer and immune system conditions
• Post-AGI: taste, original thinking, and emotional connection become more valuable, not less
Why it matters:
Hassabis built AlphaFold and AlphaGo. When he names a year for AGI, it is not a think-tank forecast -- it is an engineering roadmap from someone actively building toward it. If the timeline holds, the window to adapt -- professionally, institutionally, personally -- is four years. That is not much runway.
AI + MEDICINE
| 🔬 |
The FDA Just Fast-Tracked 10 Drugs Designed Entirely by AI |
The FDA has selected 10 AI-designed drug candidates for an expedited Phase I human trial pilot -- the first time the agency has created a formal fast-track category specifically for compounds discovered by AI. Out of 200+ AI-originated programs currently in clinical development, these 10 cleared the bar for transparent and reproducible AI discovery processes.
AI-discovered molecules have shown an 80-90% success rate in Phase I trials so far, compared to roughly 50-60% for traditionally designed drugs. The FDA published its AI Drug Development Guidance framework in January 2026, and this pilot is the first application of that framework at scale.
The numbers:
• 10 candidates selected from 200+ AI drug programs currently in trials
• 80-90% Phase I success rate for AI-designed drugs vs. 50-60% historical average
• FDA framework for AI drug evaluation published January 2026 -- this pilot is its first real test
Why it matters:
Traditional drug development takes 10-15 years from discovery to approval. If these 10 candidates progress and clear Phase I at the same rate AI-designed drugs have so far, it sets a precedent that could cut that timeline significantly. The FDA is not just approving drugs -- it is approving a new discovery method.
| 📦 |
Dropbox's Founder Is Leaving to Build AI. He Is 43. |
Drew Houston started Dropbox at 24 after losing a USB stick on a bus to MIT. Nineteen years later, he built it to 18 million paying customers and $2 billion in annual revenue. On May 26 he announced he is stepping down as CEO, moving to executive chairman, and handing the reins to product chief Ashraf Alkarmi -- who will eventually run the company alone.
Houston said he plans to start something new in AI. He noted there is "never a perfect time" for this kind of transition, but added there has "never been a more exciting period to be building things."
WHY IT MATTERS:
When a founder worth over $2 billion walks away from a profitable company to chase AI, it says something about where the next decade of value gets built. Houston is the third major tech founder this year to make this move.
| ⚖️ |
Stanford: AI Hiring Tools Screen Out Black Applicants at 10.62% of Positions |
Researchers from Stanford, Chapman University, and Northeastern studied 4 million job applications across 156 employers -- all screened by algorithms from the same vendor, Pymetrics. They found "clear racial disparities": 10.62% of positions showed adverse impact against Black applicants, and 5.32% against Asian applicants.
The compounding problem: 42 screening models are shared across employers. A rejection at one company can trigger a rejection at every company running the same model -- without either the applicant or the employer knowing why.
WHY IT MATTERS:
The data covers 2018-2022 and today's LLM-era hiring tools work differently. But the mechanism -- shared infrastructure turning individual bias into systemic exclusion -- applies to any shared model in any domain. This is what AI monoculture risk looks like in practice.
|
⚡ Coming in Hot AI Tools of the Week |
May 27, 2026 |
4 tools worth adding to your workflow right now.
| 🏄 |
AI CODING Windsurf Windsurf's Cascade mode plans and executes multi-file code changes while you watch -- think agentic coding without the back-and-forth. It was the #1 ranked new AI IDE in early 2026. Free tier includes real model access, not a watered-down demo. Try Windsurf free → |
| 🎨 |
AI DESIGN Freepik Image generation plus 200 million stock assets in one place. The Mystic model handles product shots and lifestyle photography convincingly. Useful for content teams who need generated images that actually look on-brand, not generic AI output. Try Freepik → |
| 🎙️ |
AI MEETINGS Fireflies Joins your calls, transcribes everything, and sends an AI-generated summary with action items before the meeting window has even closed. Free tier is genuinely usable -- no credit card, no meeting length limit on the core transcription feature. Try Fireflies free → |
| 💸 |
AI FINANCE Cleo An AI money coach that connects to your bank account and talks back in plain language -- not dashboards. It tells you exactly what you spent last month and where the patterns are. 7 million users. Spots forgotten subscriptions, flags spending by category, and sets savings goals through chat. Free to start. Discover Cleo → |
⚡ Quick Hits
⚡ OpenRouter raises $113M at a $1.3B valuation
The model routing API just closed a CapitalG-led Series B. It now processes 25 trillion tokens per week -- up from 5 trillion six months ago -- and has 8 million developers on the platform. See the breakdown →
🎵 ElevenLabs ships Music v2
Better vocals, richer instrumentation, mid-track genre transitions, and multilingual support. Trained on licensed data, so every track is cleared for commercial use -- a significant change from v1. Read the release notes →
💧 Google's SynthID watermarks reach OpenAI, ElevenLabs, and Kakao
100 billion+ files already carry the invisible SynthID signal. Adding OpenAI and ElevenLabs to the coalition starts to make AI-origin watermarking a cross-platform standard rather than a Google-only experiment. Learn how SynthID works →
That's all for this edition! What tools are you most excited to try?
Reply to this email - we read every response and use your feedback to improve future editions.
See you on Saturday,
The AI Tool Discovery Team
Feedback
What'd you think of today's newsletter?
Vote below to let us know how we're doing.