Good morning! Here's what's happening in AI today:
xAI's Grok rolls out Imagine Image 2.0 with precision editing and crisp text
Anthropic ships four brand new updates to Claude Managed Agents.
OpenAI's GPT-Live now supports Files and Projects for organizing real tasks
How to let Claude read your TradingView chart and write your pine script for you
4 new AI tools worth trying today
AI TOOL
xAI unveiled Imagine Image 2.0, its next-generation image model built for precision editing, crisp text rendering, and improved factuality, positioning it as a tool for real work rather than novelty generation.
Precision editing lets users make targeted changes to existing images instead of regenerating the whole scene from scratch, a common weak point in earlier image models.
Crisp text rendering addresses one of the most persistent failures in AI image generation, where text inside images typically comes out garbled or illegible.
xAI frames the model around real-world usefulness rather than pure creative output, suggesting a push toward practical use cases like mockups, ads, and documents.
Text rendering and precision editing are the two areas where AI image tools have consistently underperformed compared to what people actually need for work. If Imagine Image 2.0 delivers on both, it closes a gap that's pushed a lot of practical image tasks back toward manual design tools. The "real work" framing signals xAI is chasing utility over spectacle, a shift worth watching given how crowded the image model space has become.
AI AGENT
Anthropic shipped four updates to Claude Managed Agents, letting developers set session spend caps, pin inference to a specific geography, auto-load skills from attached repositories, and give a working agent a stronger model to consult mid-session.
Session budgets let developers set a hard spend cap when creating a session, pausing the agent with a budget_reached event once the limit hits rather than letting costs run unchecked.
The new inference_geo setting pins where a model runs, choosing between global capacity at the standard rate or in-region US inference billed at 1.1x.
Managed Agent sessions now automatically pick up skills already stored in a repository's .claude/skills/ folder, the same location used for Claude Code, removing a manual setup step.
These are the kinds of controls that separate a demo from something a team can actually run in production. Predictable spend caps and region pinning matter directly for cost and compliance, while the advisor feature gives a lighter working model a way to escalate hard calls to a stronger one without swapping the whole session. Together, they make Managed Agents meaningfully more usable for real deployments rather than one-off experiments.
AI TOOL
OpenAI updated GPT-Live to support attaching files and using Projects, letting people upload documents like a resume or itinerary and ask questions grounded in them, or organize ongoing work like a job search, trip planning, or a fitness routine inside a persistent space.
Users can now attach files like resumes or planning documents directly in GPT-Live and ask questions grounded in that specific content, rather than relying on general knowledge alone.
Projects give GPT-Live a persistent workspace, so ongoing efforts like a job search or travel plan stay organized across multiple sessions instead of scattering across separate chats.
The update targets everyday personal use cases rather than developer workflows, aiming GPT-Live squarely at how regular people already try to use AI assistants.
This closes a real gap for anyone using GPT-Live for anything beyond a single quick question. Persistent context through Projects means people no longer have to re-explain their situation every time they open a new chat, and file support means GPT-Live can actually reason about a person's real documents instead of generic advice. It's a small update on paper but changes how usable the tool is for sustained, personal tasks.
HOW TO AI
You've probably had TradingView open in one tab and Claude in another, describing what you're looking at instead of just letting Claude look at it directly.
There's a real way to close that gap. It's not an official TradingView feature and Anthropic didn't build it, it's a community-built set of MCP tools that lets Claude read your live chart, indicators, and price levels, and write Pine Script with you.
What it can't do, and shouldn't try to do, is tell you what to trade.
Get it installed and running
git clone https://github.com/tradesdontlie/tradingview-mcp.git
cd tradingview-mcp
npm installOr hand the whole setup to Claude Code in one message:
Install the TradingView MCP server. Clone
https://github.com/tradesdontlie/tradingview-mcp.git, run npm install,
add it to my MCP config at ~/.claude/.mcp.json, and launch TradingView
with the debug port. Then verify the connection with tv_health_check.TradingView Desktop needs to be running with its debug port open before Claude can see anything, the repo ships a launch script for Mac, Windows, and Linux that handles that for you.
Read the chart, then let it write your Pine Script
Once connected, plain English is enough. "What's on my chart?" pulls the symbol, timeframe, and every visible indicator. "Give me a full analysis" chains the quote, indicators, price levels, and a screenshot into one description.
This is where the tools are genuinely strongest. Describe an indicator or strategy in plain English, and Claude writes the Pine Script, injects it into the editor, compiles it, and reads back any errors, all without touching the TradingView UI.
Write a Pine Script that plots a 20 and 50 EMA, and adds a label
whenever the 20 crosses above the 50.Claude writes it, compiles it, and fixes its own errors if the first version doesn't pass. That loop, write, compile, read the error, fix, is the actual time saver.
Everything past this stays your judgment, same as it was before you connected anything. This closes the gap between staring at a chart and describing it, not between you and a decision. Read the full guide.
P.S. You can access all my prompts, workflows and AI trainings if you upgrade here
Replit shared three updates this week: security scans that run while you build, the ability to add Single Sign-On to Replit apps, and support for moving projects between workspaces.
Vercel announced that Hermes Agent can now run using Vercel AI Gateway and Vercel Sandbox, reaching every model through one gateway key with full spend observability while executing each command inside its own isolated microVM.
DeepSeek launched DeepSeek-V4-Flash-0731-Fast, now live on OpenRouter and available through multiple inference providers with early benchmarks showing sub-2-second latency and pricing starting around $0.28 per million input tokens.
🖥️ Hermes Agent: HUD mode turns it into an overlay layer on the app you're working in, drag it anywhere, ask it anything.
✨ Sugar High v2: Small, composable syntax highlighting built to be agent-friendly, 1.74KB core.
📊 Radar Researcher: Ask Internet traffic and trend questions in plain language, browser agents can now navigate Radar directly via WebMCP.
⚔️ Fleet: Run agent evals at scale in the cloud, one YAML races every pack × agent × model in parallel sandboxes, scored live on a matrix.
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