🧠 Your Run Data Just Got Smarter, Codex Builds Apps Now, and AI That Stays Private

Strava, the fitness tracking app used by over 195 million people across 185 countries, has launched an official MCP connector. The connector gives Claude read-only access to your activity data — heart rate, pace, GPS routes, cycling power metrics, clubs, and events. It connects via OAuth, the same secure authorization flow most third-party apps use, and you can revoke access from Strava settings at any time.

Before this, anyone who wanted to analyze years of Strava data with an AI had to manually export it, paste it into a chat window every single time, and hope the context window was big enough. This removes all of that.

If you're a Strava subscriber, check your MCP settings or ask Claude directly: "Am I eligible for the Strava MCP?"

OpenAI's Codex started as an AI coding agent — it wrote and fixed code, and developers loved it. This week, OpenAI announced a major expansion that pushes Codex well beyond its coding roots.

The headline feature is Sites: Codex can now create and deploy interactive, hosted websites and apps directly from a plain language prompt. You describe what you want — a dashboard, a project board, a lightweight internal tool — and Codex builds and hosts it with a shareable URL. No separate deployment setup needed. It's currently in preview for Business and Enterprise plans.

Beyond Sites, OpenAI added Annotations (a surgical in-place editing tool so you can fix one section without regenerating the whole document) and six role-specific plugins that connect Codex to 62 business apps including Snowflake, Figma, and Salesforce, with 110 automated skills built in.

Perplexity's Computer is an AI agent product that can take on complex, multi-step tasks on your behalf, using a combination of AI models. The new feature Perplexity announced is called hybrid agentic inference. The idea is straightforward: instead of routing every task to the cloud, Computer will be able to split work between a local model running on your own machine and frontier cloud models, automatically deciding in real time which parts stay local. Sensitive data — financial records, personal files, health information — stays on your device. The heavier reasoning tasks go to the cloud.

Perplexity is making the case that you don't have to choose between powerful AI and keeping your data to yourself. The feature is coming soon and will arrive first on Windows laptops.

Most website owners have Microsoft Clarity installed and check it maybe twice a month. They watch a few session recordings, feel vaguely anxious about the rage clicks, and then close the tab without doing anything. The data is there. The insight never quite arrives.

This guide walks you through connecting Microsoft Clarity to Claude and using it to run a proper behavior analysis on any page that isn't converting the way it should. You feed Claude the page context and your conversion goal, and it surfaces what's breaking — rage clicks, dead clicks, scroll dropoff, quick bounces — translated into plain English with a prioritized list of what to fix first.

Best for website owners, product managers, and marketers who are getting traffic but not getting results, and have at least a week or two of Clarity data sitting there waiting to be useful. Setup takes 15 to 20 minutes the first time. Every analysis run after that takes under 15.

✏️ Full playbook here

Tired of seeing the word "AI" slapped on literally everything?

Enshittifier is a Mac app and Chrome extension by Wells Riley that replaces every instance of "AI" in your browser with the poop emoji. That's the whole product.

It started as a joke — can you use font ligatures to turn "AI" into 💩? Turns out, yes. Riley describes it as "mostly poking fun" but also a small nudge toward being more mindful about how relentlessly the word gets thrown around. Ironically, he used AI to build it.

It won't change your workflow. But it might change your mood.

You've probably noticed that AI makes individuals faster but teams still feel just as chaotic. Miro's newly reimagined Canvas is designed specifically for that gap — the point where individual AI output needs to become collective team progress.

Miro Canvas brings human-to-human, human-to-AI, and agent-to-agent collaboration onto a single shared surface. In practice: an AI tool can generate design concepts, your team annotates and debates them directly on the board, and then AI picks back up to synthesize the feedback into a requirements doc or prototype. The canvas also connects to tools like Slack, GitHub, and Amplitude so context flows in both directions. Best for product and design teams who are already using AI individually but losing the thread when they try to work together.

HyperFrames is an open-source framework (by HeyGen) that lets AI agents create and edit videos by writing HTML, CSS, and JavaScript. When you pair it with Claude Code, you can describe edits in plain language and Claude handles the rest. Here's how to get started:

This is still early and works best for straightforward edits — captions, text overlays, illustrated diagrams. If you're newer to content creation and want clean, quick edits while you develop your own style, it's worth testing.

A full step-by-step tutorial may be coming on YouTube — let us know if you want that.