Agent Pulse
Posts
#34: Psst, Meta’s 44 Minds, Grok’s Split Personality, and Perplexity’s Browser Play

#34: Psst, Meta’s 44 Minds, Grok’s Split Personality, and Perplexity’s Browser Play

AgentPulse: Your Weekly Dose of AI Agents News.

O P
July 21, 2025

In partnership with

Welcome back, AI Agent Enthusiast!

In today’s Agent Pulse:

📢 Top Headlines
⚔️ Agent Arena
✨ Featured Agents
🎓 Free courses
📚 Must-Read Researches

You read, we listen. Got feedback? Just hit reply - we’d love to hear from you.

Enjoying Agent Pulse?
Go ad-free and support our work for just $5/month. Upgrade to Premium →

📢 TOP Headlines

Grok Miniaturizes: From Super-Grok to Baby Grok

Elon Musk’s xAI is quietly remolding Grok into a multi-tiered assistant platform layering unique personas to cater to varied user needs without calling it an “agent stack.”

Grok Assistants

xAI has introduced modular Grok personas like “Coder” and “News Analyzer” directly inside X no switching apps or plugin installs. These aren’t separate GPT clones but adaptive behavior modes triggered by context, aiming for fast, persona-driven UX that stays within the core chat dialog.

Baby Grok

Responding to backlash over Grok’s NSFW responses and extremist content, Musk announced Baby Grok, a child-safe version tailored for educational storytelling and light conversation inside X. No launch date yet, but the design is focused on filtered tone, simple Q&A, and emotional safety.

We’re going to make Baby Grok @xai, an app dedicated to kid-friendly content
— Elon Musk (@elonmusk)
2:46 AM • Jul 20, 2025

Why It Matters

Persona-first without friction: Instead of users picking tools, the chat environment adapts dynamically. That removes agent selection friction and increases usage stickiness.
Ethics in the loop: Launching Baby Grok signals xAI’s understanding that content moderation must be persona-specific, and that one-size-fits-all AI is untenable.
Platform edge: By hosting these assistants inside X, xAI keeps users locked into a native experience steering clear of App Store friction or web loading delays.

Takeaway:
xAI is revealing the next stage of agent design: contextual personas embedded in existing platforms, not standalone apps or browser plugins. That may be what real-world agent adoption looks like.

Perplexity Comet Browser: First-Mover in the AI-Powered Web

Perplexity launched Comet, a Chromium-based browser with embedded AI. Subscribers can get page summaries, agentic multi-tab context, and even task execution, booking flights, sending emails, accepting invites, all via an assistant sidebar.

The unsaid but critical implications:

Browser as agent host. Comet transforms the browser from a passive viewer into an active agent shell, offloading context, continuity, and task actions from web devs to AI.
Privacy-by-design is a competitive weapon. With local processing modes and per-tab strictness, Comet bets on trust, not extraction, a move that may define B2B adoption.
Platform arms race initiated. Meta, Google, OpenAI - they see the strategic value. Whoever owns the browser surface owns the context engine.

Agent takeaway: If your agent doesn’t own the user’s context, but lives in the browser, you’re playing catch-up. Comet signals the future browser will be the agent’s UI, not the LLM.

Meta’s “List of 44”: Building an AI Strike Team

A leaked roster of 44 researchers now working in Meta’s Superintelligence Labs reveals a powerhouse team - 50% from China, 75% hold PhDs, and 40% are ex-OpenAI, with the rest split between DeepMind, Scale, Apple, Anthropic, etc. Many earn $10M–$100M/year.

Why this hire list matters now:

Talent architecture > model architecture: Meta isn’t just acquiring minds - it's acquiring preformed research units with shared workflows and deep trust.
Geo-educated globalism: Half the team are Chinese-born and Western-educated, Meta is constructing a geopolitical blanket that buffers global AI access risks.
Big-money mission lock: The pay is massive, but retention hinges on purpose. Meta’s next move: articulate how AGI fits into a broader future for reality platforms.

This isn’t just an army - it’s a fully formed platoon with mission-blind trust. Builders should note: it’s harder to replicate just a model than to replicate a mission-driven team.

🚨 BREAKING: Detailed list of all 44 people in Meta's Superintelligence team.
— 50% from China
— 75% have PhDs, 70% Researchers
— 40% from OpenAI, 20% DeepMind, 15% Scale
— 20% L8+ level
— 75% 1st gen immigrants
Each of these people are likely getting paid $10-$100M/yr.
— Deedy (@deedydas)
3:45 PM • Jul 19, 2025