Agentic AI

ChatGPT Pulse: OpenIA's proactive AI assistant

This past week, OpenAI launched ChatGPT Pulse, a feature that conducts overnight research based on your chat history and (optionally) connected apps (such as Gmail, Google Calendar), then delivers personalized morning briefings. Currently exclusive to Pro subscribers ($200/month), Pulse represents OpenAI's latest push toward agentic AI (which works autonomously in the background). The feature analyzes your conversations, calendar events, and preferences to surface relevant updates like project follow-ups, dinner recommendations, or reminders about upcoming meetings.

Pulse requires that users have memory features turned on in ChatGPT, a feature that not all users prefer. Additionally, early testing indicates the feature becomes most useful after users explicitly indicate to ChatGPT what they want to see.

Announcement: Introducing ChatGPT Pulse
⚡ Quick Read (4 minutes)

OpenAI really, really wants you to start your day with ChatGPT Pulse
⚡ Quick Read (4 minutes)

Agentic AI

Notion 3.0 brings AI agents to your workspace

With its version 3.0, Notion just launched what might be the most accessible AI agent platform to date. The game-changer here isn't just the AI capability; it's how Notion has made agent creation accessible through natural language, letting any knowledge worker become an "AI systems architect" without technical expertise. Notion 3.0's AI agents can create documents, build databases, search across connected tools, and execute multi-step workflows autonomously for up to 20 minutes. Custom agents that work on autopilot are coming soon, promising specialized AI assistants that can handle everything from compiling user feedback to triaging IT requests.

In his paid newsletter, Nate Jones put Notion's AI agent functionality to the test. He found that success heavily depends on structured, specific prompting (casual instructions rarely work well). Still, with proper prompting, he was able to achieve some impressive results, including a Product Hub page that pulls tasks and status from meeting notes and PRDs — a page Nate created in two minutes, which he stated would have otherwise taken him a couple of days.

Introducing Notion 3.0
⚡ Quick Read (4 minutes)

AI Content Creation

Meta launches Vibes: an all-AI feed with VR energy

This past week, Meta quietly launched Vibes, a dedicated feed of short, AI-generated videos inside the Meta AI app. The feed combines MidJourney-style aesthetics with movement and music to create content that doesn't pretend to be real. Critics immediately panned it as more “AI slop,” but Ben Thompson, in his paid newsletter, argues Vibes is notable on two fronts: (1) by making the feed explicitly 100% AI, it removes the constant “is this real?” cognitive tax that plagues mixed feeds; (2) aesthetically and conceptually, it feels like Meta’s first real VR product (without a headset) hinting at a future where users browse richly imagined, unreal worlds rather than watch video clips shot by humans.

Ben is somewhat alone in his enthusiasm for Vibes (he even recorded an "emergency" podcast about it), and he does acknowledge the concerns:

"There are, needless to say, reasons to be concerned about Vibes. While other AI products can point to productivity gains or explicit positive impacts on the real world, the virtual nature of Vibes is its own condemnation: if you thought that short form video of other humans was addicting, what if we remove the last vestiges of the real world and just make a closed loop?"

He also makes the point that "Meta...more than any other company, seeks to serve people’s revealed preferences, not their stated ones, for better or for worse — and is fairly impervious to criticism along the way."

Regardless of how Vibes pans out in the marketplace, it seems an inevitable experiment and is worth checking out from a user experience perspective.

Introducing Vibes: A New Way to Discover and Create AI Videos
⚡ Quick Read (2 minutes)

Late breaking: OpenAI is introducing a similar social app in conjunction with the release of Sora 2.

Frontier Models

Chatbots are built to please, and why that's a problem

This New York Times piece explores how mainstream chatbots often reflect our views back to us, agreeing readily and reinforcing our confidence, even when we’re wrong. OpenAI has attempted to mitigate this, but it still persists. Research shows that this constant affirmation can lead to "social deskilling," eroding our tolerance for disagreement and potentially replacing the type of real-world relationships that help us grow. (In one survey, 52% of teenagers reported using AI for companionship regularly, with 20% spending as much or more time with AI than real friends.)

The article offers practical strategies to avoid the flattery trap (getting flattered into bad decisions): present questions as if asking "for a friend" to bypass the agreement bias, actively challenge the AI's responses, and remember AI chatbots are tools, not companions.

A.I. Chatbots Are Built to Please. Here’s How You Can Use Them Safely
⚡ Quick Read (5 minutes)

That’s it for this week.

Thanks for reading, and see you next Wednesday with more curated AI/UX news and insights. 👋

All the best, Heidi