Good morning! Here's what's happening in AI today:
Google rolls out Gemini 3.7 Flash, its most intelligent workhorse model for coding
Claude Code adds an auto-continue checkbox for when you hit your usage limit
DeepSeek launches V4-Pro with major agent upgrades and flexible reasoning effort
How to turn Claude Code into a research engine that thinks for free on Google's infrastructure.
4 new AI tools worth trying today
AI MODEL
Google introduced Gemini 3.7 Flash, its most intelligent workhorse model yet for coding and agents, bringing substantial gains across software engineering, web development, and complex knowledge work at an introductory price of $0.75 per million input tokens through the end of the year.
Gemini 3.7 Flash delivers substantial gains specifically in software engineering and web development tasks, positioning it as a working model for developers rather than a general-purpose chat upgrade.
The introductory pricing of $0.75 per million input tokens and $3.75 per million output tokens runs through the end of the year, making it notably cheaper to scale than typical frontier-tier pricing.
Google is explicitly framing this release around production-ready agents, suggesting the model is built for developers scaling real deployments rather than experimentation.
A workhorse model priced this aggressively changes the cost calculus for anyone running agents at scale, since the price difference compounds quickly across high-volume production workloads. Targeting coding and complex knowledge work specifically, rather than positioning this as a general Gemini upgrade, signals Google sees developer and agent workflows as the primary battleground right now.
AI TOOL
Anthropic added an auto-continue checkbox to Claude Code desktop, letting a session automatically pick back up where it left off once a usage limit resets instead of requiring a manual restart.
Turning on auto-continue means a task interrupted by a usage limit resumes on its own once the limit resets, rather than sitting idle until someone manually restarts it.
The feature specifically targets the mid-task interruption problem, where hitting a limit halfway through a multi-step job previously meant losing momentum until a person came back to restart it.
This is a small, targeted fix rather than a broader Claude Code overhaul, addressing one specific friction point in long-running agent sessions.
Long-running coding sessions that hit a usage wall used to just stop, leaving whoever was running the task to notice and manually pick it back up. Auto-continue removes that babysitting requirement entirely, which matters most for anyone running Claude Code on genuinely long, multi-step tasks that span a usage reset. It's a minor feature on paper, but it directly targets the exact moment agentic workflows are most likely to stall.
AI MODEL
DeepSeek launched DeepSeek-V4-Pro, bringing major agent upgrades with strong production gains, flexible reasoning effort across V4-Pro and V4-Flash, and native OpenAI Responses API support optimized for Codex with one-click setup, now live on app, web, and via API.
Flexible reasoning effort lets developers dial the model between low for simple tasks, high for daily agent workflows, and max for complex tasks, rather than running everything at a single fixed reasoning level.
Native OpenAI Responses API support with one-click Codex setup lowers the integration barrier for developers already building on OpenAI's API conventions, letting them adopt DeepSeek without rewriting their existing setup.
The model is available immediately across app, web, and API, with V4-Pro accessible directly through "Expert Mode" for anyone wanting to try it without a separate integration.
Adjustable reasoning effort matters directly for cost and latency, since developers can now scale compute up only for tasks that actually need it rather than paying premium reasoning costs on every simple request. Native Responses API compatibility is a deliberate move to lower switching costs for developers already invested in OpenAI's tooling conventions, effectively inviting a direct comparison on the same integration surface.
HOW TO AI
Ask Claude Code to research something with plain web search, and you're leaving most of the power on the table. It searches, it summarizes, it burns through your context doing it, and what you get back is one pass over the material, gone the moment the conversation ends.
NotebookLM is Google's free research tool, built to read a pile of real sources and answer only from what's actually in them. Bridge it to Claude Code, and Claude Code stops doing the thinking itself. It searches, it loads sources into NotebookLM, and lets Google's infrastructure do the actual analysis, for free, while Claude Code's own tokens barely move.
Meet the pipeline
Claude Code doesn't need to be the one reading fifty pages of source material. It needs to be the one orchestrating, find the sources, hand them to something built specifically for deep reading, then collect the output.
notebooklm-py, an open-source, MIT-licensed project, closes NotebookLM's API gap with an unofficial Python API and CLI, plus a ready-made skill for Claude Code, Codex, and OpenClaw.
Bridge Claude Code to NotebookLM
uv tool install "notebooklm-py[browser]"
notebooklm login
notebooklm auth check --test --json
notebooklm skill installLoad sources straight from the terminal or a Claude Code conversation:
notebooklm create "My Research"
notebooklm use <notebook_id>
notebooklm source add "https://en.wikipedia.org/wiki/Artificial_intelligence"
notebooklm source add "./paper.pdf"Let NotebookLM do the thinking
notebooklm ask "What are the key themes across these sources?"
Run this from inside Claude Code and the difference is immediate: Claude Code's own context barely grows, because the heavy reading happened on Google's side, not in its own conversation window.
Automate the whole loop from one prompt
Use the yt-search skill to find the latest trending videos on
[your topic]. Once we have those, send them to NotebookLM using
the notebooklm skill. Give me its analysis on the top points,
then have NotebookLM create a portrait-style infographic
depicting that analysis.Claude Code searches, filters results by recency, loads the URLs as sources, asks NotebookLM for the analysis, and generates the infographic, all from one message.
P.S. You can access all the AI trainings, prompts and workflows if you upgrade.
OpenAI launched Computer History in the ChatGPT desktop app, letting ChatGPT remember a person's activity across apps and websites on their computer so future interactions feel more personalized and require less explanation.
Z.ai introduced GLM-5.3, built for top-tier coding and agentic capabilities through post-training on the 743B base model, marking what the company calls a major leap in cybersecurity performance among open models.
Cursor shipped a performance upgrade that makes cloud agents start 3x faster, powered by builds, ready-to-use development environments Cursor prepares continuously in the background at no additional cost.
🎵 MiniMax: MiniMax-Music3, next-generation open-weights music model, production-ready and versatile.
🔍 Firecrawl: 41M+ life science papers added to the Research Index, drug discovery and clinical trial search at 90% recall, live in the API.
🎬 ChatCut: ChatCut Desktop, AI video editing that's now local, fast, and free, connect Codex or Claude Code in one click.
🎙️ Sarvam: Voice Agents now available to everyone, human-like agents that remember context across conversations.
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