🧠 Open Models, Tax AI, and ChatGPT in Your Car

 🔥🔥🔥 Three Highly Curated AI Stories

1. Google just open-sourced a Gemini-level model

Google released Gemma 4, its most capable open model family to date, built from the same research that powers Gemini 3. They shipped it under an Apache 2.0 license, which means you can run it on your own hardware, build commercial products with it, and do basically anything you want.

Here's why that's a big deal. Advanced reasoning and agentic AI (AI that takes multi-step actions on your behalf, rather than just answering questions) has mostly lived behind closed APIs, with usage limits and monthly bills attached. Gemma 4 brings that capability to your own machine, free.

Fair caveat: you still need reasonably powerful hardware to run these models well. But for developers, researchers, and builders, the gap between "frontier AI" and "AI I actually own" just got a whole lot narrower.

2. Perplexity Computer will now help you file your taxes

Perplexity is best known as an AI-powered search engine. But it has a feature called Computer that lets it browse the web and take real actions on your behalf — wait, no em dashes. Let me rephrase. Perplexity Computer goes beyond search; it can browse and act. This week, they pointed it at federal tax preparation.

Select "Navigate my taxes" inside Perplexity Computer and it'll walk you through filing your federal return, reading your situation and pulling relevant information as it goes.

The obvious caveat: this is not a licensed tax professional, and you should verify anything before you submit. But step back and look at the bigger picture. AI is now doing the kind of work people spend hundreds of dollars hiring professionals to handle. The category of "tasks only humans should touch" is getting smaller by the month.

3. ChatGPT is now in your car via CarPlay

OpenAI announced this week that ChatGPT is now available through Apple CarPlay. If you're running iOS 26.4 or newer and connected to a supported car, you can start a voice conversation with ChatGPT directly from the dashboard interface or pick up where you left off from your phone.

The timing makes sense. Voice AI is the obvious interface for driving, and hands-free AI is still underused relative to how much time people spend commuting. The practical upshot: your drive to work is now a window to brief yourself, brainstorm, or get answers to the questions stacking up in your head. Rolling out now.

🔥🔥 Two Pro AI Tools To Try Today

1. 🗂️ MCP.Directory

MCP stands for Model Context Protocol, and it's essentially the system that lets AI tools like Claude, Cursor, or ChatGPT connect to external apps and services. Think of it like an app store, but for AI integrations. MCP.Directory is the largest searchable index of these connectors, currently listing 3,000+ MCP servers covering everything from Gmail and Google Drive to Snowflake, GitHub, and Notion. You can filter by category, see which AI clients each server supports, and install most of them in one click. If you're trying to build an AI that actually knows your workflow, this is where you start.

2. 🎓 Anthropic Academy (Skilljar)

Anthropic just opened a free learning platform with structured courses on Claude. You can learn how to use the Claude API, build MCP servers from scratch, get hands-on with Claude Code, or pick up AI fluency fundamentals. Most courses are free, structured as video lessons with exercises, and many offer a certificate when you finish. The Building with the Claude API course alone is 8+ hours of practical content. Worth bookmarking whether you are a developer or someone trying to understand what Claude can actually do.

🔥 Things You Should Know About AI

Give Claude a permanent memory with a single markdown file

By default, Claude (and most AI tools) forget everything when a conversation ends. But there's a simple fix: a CLAUDE.md file. It's a plain-text markdown file you drop into your project folder, and Claude Code automatically reads it at the start of every session. It becomes Claude's briefing document: your preferences, project context, naming conventions, commands, and anything else you'd otherwise have to re-explain every time.

Here's how to set one up:

1. Open your project folder in Claude Code and type /init — Claude will scan your project and generate a starter CLAUDE.md file automatically.

2. Open the file and trim it down. Delete anything obvious (Claude can read your package.json; it doesn't need you to repeat it). Keep instructions that actually change Claude's behavior.

3. Add your real preferences: how you name files, which commands to run tests, any architectural decisions Claude should respect.

4. Use the # shortcut mid-session to add new rules on the fly — Claude writes them to CLAUDE.md for you.

5. Treat the file as a living document. Whenever Claude makes an error you've seen before, add a line to prevent it next time.

The result: a Claude that walks into every session already knowing how you work, without you having to explain yourself again.