Amazon Quietly Dropped a Game-Changer

Welcome back apprentices! 👋

Let’s be honest — most AI tools talk a good game but crumble the moment you ask them to actually do something useful.

But something different just landed.

No buzzwords, no sci-fi promises. Just a new kind of AI that quietly gets to work in the background — clicking buttons, filling out forms, and handling all those little tasks that normally eat up your day.

It’s not trying to sound smart. It’s trying to be helpful. And it might be the most practical thing Amazon’s launched in years.

In today's email

• Brain-to-Voice Breakthrough

• The Drama Behind Altman's Firing

• DeepMind Puts a Lock on Its Lab

• Meet Nova Ac

• Therapy Bots Are In

• Even more AI magic

• Test the prompt (NEW)

_{Read Time: 4 minutes}

Quick News 

🧠 Scientists at UC Berkeley and UCSF just built a brain-to-speech device that’s basically a telepathic Siri with manners. The system takes brain signals from people who can’t speak and — get this  —streams their thoughts into fluent, natural-sounding speech in under a second. No more robotic lag or 8-second loading bars for every sentence. It even recreates the user's own voice, making it feel more like “them” and less like a text-to-speech GPS from 2007.

🔥 A spicy new excerpt from The Optimist reveals the real tea behind Sam Altman’s 2023 OpenAI firing, and it’s more Succession than Silicon Valley. Turns out, co-founder Ilya Sutskever and then-CTO Mira Murati assembled a dossier of alleged toxic behavior and shady fund ownership (Altman reportedly owned OpenAI’s Startup Fund—surprise!). They handed it to the board, pulled the trigger... and accidentally triggered a company-wide mutiny. The result? Altman was back in his chair by Monday, and Sutskever and Murati were out like a failed GPT prompt.

🔐 DeepMind, once the golden child of open AI research, is now hoarding its homework like it’s finals week. Papers that might tip off competitors—or reveal flaws—are getting delayed up to six months, with multiple layers of approval. Turns out, when you’re in an AI arms race, “peer-reviewed” starts looking more like “profit-reviewed.” With Gemini, Astra, and other shiny products rolling out, the lab is pivoting hard toward commercial wins. For the rest of us, that means fewer breadcrumbs to follow and more mystery meat under the AI hood.

Amazon 
Alexa’s Nerdy Cousin Just Got a Job in IT 

In the battle of AI agents, Amazon just showed up fashionably late… and crushed the party. Meet Nova Act, the new AI agent system from Amazon AGI Labs (yes, that AGI), purpose-built to take actions inside your web browser like it’s doing its taxes.

Think of it like this: while everyone else is building AI that can talk about doing things, Nova Act actually does the things. It fills out your forms, books your vacation days, manages your messy calendar — and never once asks, “Did you mean this?”

And it gets better. You can now build your own agents using the Nova Act SDK, which is basically a power tool for devs tired of writing 300 lines of automation scripts just to click a “Submit” button on a legacy internal dashboard.

Nova Act vs. The Browser Task Mafia

Let’s face it — most current AI agents are like interns on day one: super enthusiastic, but they’ll ask you to explain what a “dropdown menu” is and then crash Excel.

Amazon looked at this and said: “What if we just built one that worked?”

Nova Act is trained not just on language or APIs, but on actual browser environments. It understands popups, dropdowns, form fields, weird JavaScript modals, and buttons that only render once you scroll down. This isn’t just smart — it’s street smart.

Amazon didn’t come to play. It benchmarked Nova Act against the biggest names in the agent game:

🥊 And the winner is...

• Nova Act leads where it counts — UI complexity. It nails the weird stuff: pop-ups, modals, date pickers, and visual icons.
  

• Claude 3.7 Sonnet is like the agent that aces multiple-choice but freezes during open-ended questions.
  

• OpenAI’s Computer Use Agent is reliable-ish, but can trip over a random tooltip and call it a day.
  

And here's the kicker: Nova Act doesn’t just do better on benchmarks — it comes with tools. 

Real, grown-up tools like:

• 🎮 Playwright integration for direct browser control
  

• 🧠 Python threading & test hooks
  
  

• 💻 Atomic commands to break down complex tasks
  

• 🕶️ Headless mode for when you want to schedule it to do stuff while you’re sipping margaritas
  

It’s the agent you’d actually hire.

Under the Hood

Amazon's SDK is a dev’s dream. You can:

• Interleave code and commands like a wizard
  

• Set up complex workflows without babysitting
  

• Override agent behavior (“don’t buy the upsell, please”)
  

• Run it async in the background like a well-trained daemon
  

They’ve even got test agents ordering salads for Tuesday dinner. This is not a drill.

Oh, and in case you're wondering who built this? The team includes David Luan and Pieter Abbeel, both ex-OpenAI AI overlords. It’s like Amazon drafted the Avengers of AGI to make Nova Act happen.

So what’s the deal? Nova Act is Amazon’s sleeper move in the AI wars. While Google and OpenAI battle for chatbot dominance, Amazon’s out here building the first truly useful AI employee. And they’re rolling it into the upcoming Alexa+, which means this tech is about to land in millions of homes.

If it works as advertised, Nova Act will quietly become the first mass-market autonomous agent — the kind that can book your meeting, write your OOO email, and maybe, just maybe, reorder your weird oat milk from Whole Foods.

It’s not about flashy conversations anymore. It’s about doing the work — and Nova Act shows up with a lunchbox and a union card.

AI Therapy
Therapy Bots Are In — And They Don’t Charge $200 an Hour

In a first-of-its-kind trial, Dartmouth ran a study on a therapy chatbot — and surprise, it didn’t just talk back, it actually helped. Over a few weeks, users reported a 30% drop in anxiety and depression symptoms, based on legit clinical tools like the PHQ-9 and GAD-7.

Even better? People used it 4+ times per week, without ghosting it like their last real therapist. The study suggests that yes, an AI chatbot might not ask you “how that makes you feel” with a knowing nod—but it can provide actual, measurable mental health support.

AI-Driven Therapy vs. Traditional Care

Where can chatbots lend a (non-human) hand?

Here’s where AI therapy shines — and where it knows when to step back:

AI’s Sweet Spot:

• 24/7 Emotional Support: No waiting rooms. No calendar syncing. Just immediate, always-on text-based support.
  

• Mood Tracking & Check-ins: It won’t forget to ask how you’re doing (even if you pretend you're “fine”).
  

• CBT Micro-Exercises: Delivers cognitive-behavioral techniques in digestible, daily formats.
  

• Stigma Reduction: Talking to a bot feels less vulnerable for first-time help seekers.
  

• Triage at Scale: Identifies users needing higher-level care and flags serious issues early.
  

Leave It to the Humans:

• Complex Trauma Therapy: Bots don’t do breakthroughs or tissue boxes.
  

• Diagnosis & Medication Decisions: They're therapists, not pharmacists — or MDs.
  

• Crisis Intervention: If you're in an emergency, AI will reroute you — but it won’t hold your hand.
  

• Emotional Nuance: Sarcasm, dark humor, or the phrase “I’m fine” still trip up the algorithms.

Think of AI therapy like the mental health intern who never sleeps — it won’t replace your top clinician, but it’ll handle the check-ins, take notes, and maybe even save a life at 2 a.m.

What’s the Deal? Mental health care is overwhelmed, underfunded, and painfully slow. AI-powered therapy isn’t here to replace professionals — but it’s sneaking in as a shockingly capable sidekick, ready to fill gaps, manage scale, and listen to your spiral at 2 a.m. (without judging your coping playlist). 

For healthcare orgs and digital health platforms, it’s a major opportunity: low-cost, high-impact, always-on support.

Help Your Friends Level Up! 🔥

Hey, you didn’t get all this info for nothing — share it! If you know someone who’s diving into AI, help them stay in the loop with this week’s updates.

Sharing is a win-win! Send this to a friend who’s all about tech, and let’s bring them into the fold!

Even Quicker News 

📡 After Myanmar’s massive quake, Microsoft’s AI leapt into action — only to realize satellite vision doesn’t work through clouds (who knew?). Once the skies cleared, it ID’d 2,000+ wrecked buildings to help the Red Cross, proving AI can do a lot… just not weather control (yet).

🧪 Google-backed Isomorphic Labs raised a casual $600M to let AI play doctor — faster pills, fewer clinical tantrums. Pharma just got a new lab partner that doesn’t sleep or need FDA approval to think.

💨 The UK’s rolling out an AI lung test that spots serious conditions faster than you can say “inhaler,” and without needing a specialist to squint at your scan. It's like WebMD—but this time, it’s actually useful.

Today’s Toolbox

🎬 Runway’s new Gen-4 model spits out slick, 1080p videos with shockingly consistent characters and scene dynamics — basically, your VFX guy that doesn’t take lunch breaks. It’s already working on Amazon shows and Madonna’s tour, so yeah… your next video editor might not be human.

🎯 Tinder just launched The Game Game™, an icebreaker that helps you get weird before you meet IRL — because nothing says romance like answering mildly embarrassing questions with a stranger. It’s not about being right, it’s about being real… and maybe oversharing just the right amount.

🧥 H&M is swapping catwalks for code with AI-generated models designed to cut costs and boost campaign speed. It’s efficient, scalable, and trend-responsive — but creatives and labor advocates are already raising eyebrows (theirs, not the AI’s). 

🧪 Test the Prompt

A playground for your imagination (and low-key prompt skills).

Each send, we give you a customizable DALL·E prompt inspired by a real-world use case — something that could help you in your business or job if you wanted to use it that way. But it’s also just a fun creative experiment.

You tweak it, run it, and send us your favorite. We pick one winner to feature in the next issue.

Bonus: you’re secretly getting better at prompt design. 🤫

🎨 Prompt: “Personal Brand, but Make It Epic”

“Design a dramatic, over-the-top mural that represents the personal brand of a [profession or persona]. The mural is painted on the side of a [bizarre location], featuring symbolic elements like a [mythical creature], a [weird object used metaphorically], and a scene of [ridiculous success moment]. The person is in the center, wearing a [strange outfit] with glowing eyes, surrounded by fans taking selfies with [random tech gadget]. Add dramatic lighting, clouds parting, and a giant AI assistant floating above holding a sign that says [funny quote about success].”

We’ll be featuring the best generations in our next newsletter!

_FEEDBACK