Good Morning! Here's what I have for you in today's newsletter:
Google launches Nano Banana 2.1, sharper image editing at half the API price
ElevenLabs adds Architect, an expert that builds and fixes your agents
OpenAI opens the Decisions API for fast routing and classification calls
Build a one-person clipping agency in Grok Bot
4 new AI tools worth trying today
AI MODELS
Google released Nano Banana 2.1, its latest model for generating and editing images. It's available in the Gemini app, AI Mode in Search, Google AI Studio, the Gemini API, Flow, and Google Ads.
Mark the area you want to change, and the model edits that region. It can keep track of up to four people and 10 objects across edits, with up to 14 reference images per request.
API image prices are roughly halved: $0.0336 for a 1K image and $0.0756 for 4K. Input tokens, however, now cost $1.50 per million, three times the previous rate.
You can refine images through conversation, use search results as context, and adjust how much the model thinks. It's free to try in Gemini within your plan's daily image limits.
Changing a product's color shouldn't mean getting a different face or background along with it. Selecting an area gives you more control over each revision. The lower image prices also help, though long prompts and reference images add to input costs. If your app still uses Nano Banana 2, mark your calendar, you’ll need to switch before October 29.
AI AGENTS
ElevenLabs launched ElevenAgents Architect, a built-in assistant for creating and improving voice and chat agents. Tell it what you need by speaking or typing, and it builds the changes inside ElevenAgents.
Ask why customers keep requesting a human during refund calls, and Architect searches thousands of call transcripts for patterns and possible fixes.
Suggested fixes come already built and tested in simulation. Each change is saved as a versioned draft you can review or revert before publishing.
To create an agent, describe what it should do and attach any relevant files. You can also access Architect through Claude, Claude Code, ChatGPT, Cursor, and Grok Bot.
When support calls keep going wrong, finding the cause can mean hours of reading transcripts. Architect can investigate the problem and return a proposed fix for you to review. Simulation gives you a way to test the change, and approval is required before customers encounter it.
AI TOOLS
OpenAI released the Decisions API in public beta for all developers. Powered by GPT-6 Luna, it helps apps quickly choose which model, tool, or action to use next.
OpenAI reports decisions up to 10 times faster than calling GPT-6 Luna through the Responses API. Both text and image inputs are supported.
Predicates estimate how likely a statement is to be true. Choices pick from a list of options and return a confidence score for each.
Scores rate an input on a numeric scale, letting you sort support tickets by urgency or rank content by how relevant it is.
An app can waste time deciding where to send a request before it even starts answering. This API speeds up that first step, such as choosing a cheaper model for a simple question or a stronger one for harder work. Confidence scores also let you set a cutoff for human review.
HOW TO AI
Three bots for clips, thumbnails, and post drafts. You make the final decision.
Three Grok Bot templates can help turn a long recording into a finished clip pack. This guide covers production; finding clients and setting prices are still your job. You'll need an eligible Grok Bot plan, the desktop app, and a recording you're allowed to clip.
Step 1: Choose the roles
Clip Bot finds moments and cuts approved clips. Stills & Clips Desk pulls and sizes stills. Copy Humanizer rewrites post text in your voice. Find the templates at x.ai/bot/marketplace.
Step 2: Check the templates
All your bots use the same computer, files, and logins, so read each template before adding it. On the template page, select Import Bot, then Add Bot in the app. Ask each bot to list its skills and routines, because a reported import bug can leave skills missing.
Step 3: Create the group
In New chat, select the three bots and name the group Clip Pack. Type @ and select a bot to give it a specific job.
Step 4: Set the handoffs
Answer each bot's setup questions in its own chat. Then paste the rules into the group: Clip Bot suggests five moments and waits for your selection, the other two bots work from approved clips, nothing publishes, and handoffs use file paths because messages between bots are text only.
Step 5: Test a clip pack
Upload a recording or paste a link, choose three moments, and check quotes against the transcript. Then open Settings, then Usage & Billing, and set the monthly on-demand limit to None or Fixed. Keep routines off until a test comes out clean.
P.S. You can access all my AI courses, trainings, prompts library, and $500+ worth of AI tool discounts here.
Anthropic brought Claude into Google Docs, Sheets, and Slides. Its sidebar reads the open file and makes edits with your approval for each change. You can also open Google files beside a chat in Claude, with existing sharing permissions still applying.
Google released EmbeddingGemma 2, its first open model built for multimodal embeddings on devices. Based on Gemma 4 and licensed under Apache 2.0, it puts images, video, audio, and code into a shared representation for comparison and search.
OpenAI published 722 math manuscripts from an unreleased internal model, covering 372 groups of related results. Many include Lean proofs, though outside mathematicians have yet to confirm any of the results.
🎞️ Tesseract: added open-source converters so Claude and ChatGPT can create and open Premiere and After Effects project files, with conversion supported in both directions.
🖼️ fal: now works inside ChatGPT and Codex, letting you browse generated images, videos, and your fal Media Library directly in the chat.
🏈 Firecrawl: added fantasy sports data to Alexandria, bringing player stats, team analysis, news, and betting odds to ChatGPT and Claude.
🔓 Mistral: released Mistral Large 4, an open-weight model built for multimodal input, with 1 trillion total parameters and 49 billion active per token.
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
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial
✓ $500+ worth of AI tools