Good Morning! Here's what I have for you in today's newsletter:
Google adds 13 Connected Apps to Gemini
ElevenLabs launches an Image and Video API with 18 models on one key
Anthropic moves Claude Code cloud sessions out of preview with free credit
Learn the five fixes that stop Opus 5.5 from wasting your tokens
4 new AI tools worth trying today
AI TOOLS
Google added 13 new Connected Apps to the Gemini app across productivity, creativity, and lifestyle, rolling out to signed-in users and enabled from settings or an @ mention.
Productivity picks are Airtable, Linear, monday.com, PandaDoc, Wispr AI, and Zoho, so a prompt can list high-priority Linear issues, query an Airtable base, or create a monday.com board item.
Creativity adds Adobe, Picsart, Squarespace, and Webflow, with examples like relighting a photo to golden hour in Adobe or adding and publishing an FAQ section on a Webflow site.
Lifestyle brings apartments.com for listings, Experian for credit monitoring, Peloton for workouts, and SeatGeek for tickets, and every connector runs on MCP, so any partner can build one.
The useful ones here are the work apps. If your team runs on Linear or Airtable, Gemini can read and update them from the chat you already have open without tab switching. Google hasn't provided everything about what action each app supports, so try the one you’re familiar with first.
AI MODELS
ElevenLabs launched an Image & Video API, so the same ElevenLabs API key and SDK developers use for speech can generate images and videos across 18 third-party models.
Video models include Google Veo 3.1 and ByteDance Seedance 2.5. Image models include OpenAI's GPT Image 2.5 and Google's Nano Banana Pro, all called through one endpoint.
Speech, transcription, dubbing, music, sound effects, images, and video share one account, one key, and one billing line, with credits deducted per generation.
The API offers the same models as ElevenCreative, so a clip generated by code can be opened in Studio for captions, voiceover, and export.
If you already pay ElevenLabs for the voice features, now you can also produce a product image or a promo clip in the same pipeline with the same key. Two things to keep in mind: ElevenLabs hasn't published per-model API pricing in the post, and check that the model you want is one of the 18.
AI AGENTS
Anthropic moved Claude Code cloud sessions out of research preview, running tasks on Anthropic-hosted infrastructure so work continues with the laptop closed, available from the web, desktop, mobile, and CLI.
Sessions start at claude.ai/code, the Code tab in the mobile app, the desktop app, or claude --cloud in the CLI, and each requires a connected GitHub account.
Existing subscribers get a one-time credit for cloud sessions only, $100 on Pro and $250 on Max, claimed from the link or with /claim-credit in the CLI.
The credit sits outside normal usage limits and applies automatically, so a session that hits a local limit can continue in the cloud until the credit runs out.
If you hit your Claude Code limit mid-task, the credit lets you finish it in the cloud without waiting for the reset. Start a long refactor or test run, close the laptop, and check the result from your phone. Claim it before October 7.
HOW TO AI
Prompts written for older Claude models carry habits that frontier models don't need anymore, and those habits cost tokens. Anthropic's own testing found removing them cut cost by 14.6% while accuracy went up too.
Step 1: Set effort to medium
Higher effort isn't always better, it can mean Claude deliberates longer than a task warrants without improving the answer.
/effort medium
Step 2: Run prompt audit
Prompts accumulate instructions that patched older models' weaknesses, verification rituals, emphasis boosters, mandatory scratchpad steps. Frontier models follow these literally, burning tokens on unnecessary work.
/claude-api prompt-auditAnthropic's own benchmark found this cut cost by 14.6% and raised accuracy by 5.3% during a model migration.
Step 3: Protect the cache
Caching reuses an identical prompt prefix instead of recomputing it, but breaking it costs 1.25x the normal price.
Check my prompt cache hit rate and tell me where misses are happeningKeep volatile values like timestamps out of your system prompt, and don't let tool definitions reorder themselves between calls.
Step 4: Stop-and-go rule
Left alone, it sometimes stops to report instead of continuing. Name the stops you want directly.
When a step doesn't need my input, keep going. Put status notes in the same message as your next action.
Stop and ask only when you can't continue without me, or before anything destructive: deleting data, force-pushing, or changing anything outside this repository.
Step 5: Log tasks to a file
A checklist in the scrollback can get lost when a long run compacts. One kept in a file survives it.
Keep a checklist in TASKS.md. Tick each item when it's done, and add anything new you find.P.S. You can access all the AI courses and trainings (including the full version of this one), prompts and workflows if you upgrade.
Perplexity brought Portable Computer for Windows to AMD Ryzen AI Max processors, so local agents can run on those machines against connected apps and local files, entirely on device.
Google added local on-device Gemma 4 to the Antigravity SDK, so developers can build multi-agent workflows pairing cloud models with Gemma 4 agents to audit, patch, and test code with no API fees and no data leaving the machine.
Firecrawl connected Alexandria to Claude through the Firecrawl connector, giving agents access to 100+ data providers and 113M+ indexed sources.
🛠️ Krea: added custom apps to Krea Agent, so describing a character studio, 3D viewer, or motion effect tool gets it built for you.
🧩 Cline: added Gemini 3.8 Flash free, scoring 41 on the Artificial Analysis index at 291 tokens per second with a 1M context window.
💬 Slack: added n8n agents, which answer support questions in-channel, trigger workflows from a message, and send digests you can reply to.
🌐 Odyssey: Agora-2 is a multi-agent world model where up to 20 humans and agents interact in one real-time environment, open as a multiplayer research preview.
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