Good morning! Here's what's happening in AI today:
Anthropic launched Claude Sonnet 5, its most agentic model yet
NotebookLM launched Short Video Overviews for any topic
Google AI Studio released Nano Banana 2 Lite for fast image generation
PLUS: How to turn Claude into a full-time AI employee in 7 days
4 new AI tools worth trying today
AI MODEL
Anthropic launched Claude Sonnet 5, its most agentic Sonnet model yet, now the default model for Free and Pro users across every plan and available immediately in Claude Code and the API.
Claude Sonnet 5 makes plans, uses tools like browsers and terminals, and runs autonomously at a level that until recently required larger and more expensive models, closing much of the gap with Opus 4.8 at Sonnet speed and price.
On agentic coding benchmarks, Sonnet 5 scores 63.2% compared to Sonnet 4.6's 58.1%, and it actually outperforms Opus 4.8 on knowledge work tasks, making it the strongest everyday model Anthropic has shipped.
Introductory pricing runs at $2 per million input tokens and $10 per million output tokens through August 31, after which it moves to $3 and $15, with prompt caching offering up to 90% cost savings on top of that.
Sonnet 5 also shows a lower rate of undesirable behaviors than Sonnet 4.6, hallucinates less, and is better at refusing malicious requests and resisting prompt injection attacks in agentic contexts.
Sonnet 5 is now the free default. Every person using Claude on a free plan woke up today with a meaningfully more capable model. For developers, a model that rivals Opus on knowledge work at roughly 60% less cost changes the build calculus entirely. This is the kind of release that shifts what people expect from the mid-tier.
AI VIDEO
NotebookLM launched Short Video Overviews, a new feature that turns any source into a 60-second vertical video that deep dives into any concept, rolling out now to Google AI Ultra and Pro subscribers on mobile and web with free users coming soon.
Short Video Overviews take the sources already in your notebook and generate compact vertical videos designed for mobile viewing, bringing the depth of a NotebookLM analysis into a format built for how people actually consume content on their phones.
The format targets complex material specifically — the framing is "doom scrolling but educational," positioning it as a way to absorb dense research or academic content in the same format and session length as a social media video.
The feature sits alongside NotebookLM's existing output formats including Audio Overviews, Cinematic Video Overviews, slide decks, and infographics, giving users more ways to turn the same sources into different content types without leaving the app.
Rolling out now to Google AI Ultra and Pro subscribers on both mobile and web, with free users getting access soon, meaning this will eventually reach NotebookLM's full audience.
NotebookLM keeps expanding what you can do with a source without making the source better. Short Video Overviews add a format that fits naturally into how people already spend time on their phones, which means research and learning can happen in moments that previously went to mindless scrolling. Free users getting access soon makes this one of the broader Google launches of the week.
AI DESIGN
Google AI Studio launched Nano Banana 2 Lite, its fastest and most cost-effective Gemini image model yet, delivering text-to-image outputs in 4 seconds at just $0.034 per 1K-resolution image via AI Studio and the Gemini API.
Nano Banana 2 Lite is built specifically for high-velocity developer pipelines, prioritizing speed and cost over maximum quality, making it the right choice for workflows that need to generate large volumes of images quickly without paying flagship model prices.
At $0.034 per 1K-resolution image and a 4-second generation time, the model gives developers a concrete cost structure to build around, removing the unpredictability that makes high-volume image generation expensive to scale.
The model is available today by swapping it into existing Gemini image workflows via AI Studio and the Gemini API, meaning developers already using Nano Banana 2 can adopt the Lite version with minimal integration changes.
The launch extends the Nano Banana 2 family with a purpose-built tier for volume use cases, giving developers a clear choice between maximum quality and maximum throughput depending on what their pipeline needs.
Image generation at scale has always required a trade-off between quality and cost. Nano Banana 2 Lite removes that tension for use cases where speed and volume matter more than frontier output quality. For developers building pipelines that generate dozens or hundreds of images daily, a 4-second model at $0.034 per image makes previously expensive workflows genuinely affordable.
PREMIUM GUIDE
Last Sunday I sat down and listed every task I do in a week that involves Claude.
Writing video scripts. Researching market trends. Drafting newsletters. Generating content ideas. Pulling competitor analysis.
Then I counted how many times I had typed basically the same prompt from scratch to do those things.
The number was embarrassing.
I wasn't using Claude as an employee. I was using it as a search engine I had to bribe every single time. Open a chat. Re-explain the context. Re-specify the format. Get an output that was close but not quite right. Close the tab. Do it again tomorrow from zero.
That's not AI working for you. That's you working for AI.
Seven days later, I had four workflows running automatically. A morning brief waiting when I opened my laptop. Content research delivered before my first coffee. Built without writing a single line of code.
Here's the exact system, day by day.
Day 1: Write the job description.
Every employee starts with one. Your AI employee is no different. Answer five questions before touching any tool: What is it responsible for? What does a perfect workday look like? What decisions can it make alone? What should it always escalate? What does good work look like? Those five answers become your system prompt. Everything else is built on top of them.
Day 2: Pick your interface.
Claude has three. Claude Chat is for one-off questions - no memory, no scheduling, no automation. Claude Cowork is for non-technical users who want to build real workflows without code. Claude Code is for developers who want full control through the terminal. If you're not a developer, start in Cowork. Download Claude Desktop and open the Cowork tab.
Day 3: Build your first workflow.
Every workflow has four parts: a trigger (what starts it), inputs (what data it needs), a process (step-by-step instructions), and an output (what the result looks like and where it goes). Build the single most time-consuming recurring task from your job description. Run it once manually to test the output. If it looks right, type /schedule and set a cadence. That workflow now runs automatically.
Day 4: Give it memory.
Create a context.md file with who you are, how you write, your quality standards, your tools, your rules, and two or three examples of work you've loved. At the start of every session, tell Claude to read it first. The more context it has, the more it performs like someone who's worked with you for years rather than someone you just met.
I use Typeless for voice-to-text, so I talked through my whole workflow into a voice note, had it transcribed, and pasted it into the file. Took 12 minutes.
Day 5: Connect your tools.
Claude supports connectors for Gmail, Google Drive, Slack, Notion, Microsoft 365, GitHub, Linear, and Zapier. Add the two or three that your Day 3 workflow actually needs. Don't add everything at once. A workflow that was reading a local file now reads your inbox. A briefing that saved to a folder now posts to a Slack channel.
Day 6: Build your routine stack.
Four workflow types cover almost everything: one daily (morning briefing), one weekly (performance synthesis), one event-triggered (fires when something specific happens), and one on-demand (you trigger manually for deep work). Each one running saves 30 minutes to 2 hours every time it fires.
Day 7: Set the review system.
Every Friday, run a review of every output produced that week. For each workflow, rate the quality from 1 to 10. If anything falls below 7, diagnose why and propose the specific prompt change that fixes it. Spend 15 minutes reading the review and approving the changes. That's the compounding loop. After four weeks, your outputs are barely recognizable compared to where they started.
I run this review before I go to the gym on Fridays. By the time I'm back, I've already read it and made my notes.
The tools are here. The playbook is here. The only variable is whether you build it.
Share this with one person still copy-pasting prompts every morning. That's all I'm asking.
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial
Anthropic launched Claude Science, a new research app available now in beta that connects to over 60 scientific databases, traces every artifact back to the code that generated it, and gives researchers on-demand computing environments to run and return to workflows without losing state.
Google AI Studio released Gemini Omni Flash, a high-quality, cost-efficient model for video generation and conversational editing that lets developers refine videos using natural language and simple prompting, available today via AI Studio and the Gemini API.
Nous Research updated Hermes Agent with a new scraping backend that makes web reading up to 60x faster and 49x cheaper, passing clean content directly to the agent without redundant processing steps.
🤖 Booster Studio: The world's first embodied AI IDE, combining code editing, high-precision simulation, real-robot debugging, and physical deployment in a single platform for anyone building physical AI.
💬 Bloome: A human-AI teammate system where Claude, ChatGPT, and DeepSeek collaborate together in a shared context with no information loss, working through drafting, challenge, and refinement until the output is right.
🎙️ Seed Audio 1.0: Higgsfield's audio model for voice cloning, text narration, and video dubbing across 18 languages, available directly on Higgsfield and on Claude via the Higgsfield MCP.
💰 ChatGPT Personal Finance: Connect your financial accounts securely, see where your money is going, and get personalized financial guidance grounded in your actual data. Now available to Plus users in the US.
THAT’S IT FOR TODAY
Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!
See you tomorrow :)
- Dr. Alvaro Cintas