Why can't we be [AI] friends?

Agent arms race.

$81T typo.

 

View in browser

March 4, 2025

Better late than never

Hi there, 

 

AI chatbots saw a record 427M app downloads last quarter — up 42% vs. Q3.

 

Now Meta wants in on the action. 

 

The tech giant reportedly plans to spin Meta AI into a standalone app to compete with ChatGPT, Google’s Gemini, Perplexity, and other chatbot apps.

consumers are flocking to ai chatbots-1

ChatGPT is still king with nearly a quarter of AI chatbot app downloads in 2024 and 400M weekly active users across platforms (not just mobile).

 

But Meta has a massive in-built user base, including ~700M people who already interact with Meta AI features each month across apps like Facebook and Instagram.

 

The battle for attention is on.

the battle is on altman

Source: X

One of the differentiators for social AI applications will be emotional intelligence — an area where OpenAI’s latest model, GPT-4.5, shows promise. 

 

While many headlines have focused on the model’s eye-watering costs, one notable advance is its reported ability to interact with greater empathy (an area where Anthropic’s Claude has so far been superior).

 

This matters because the more natural AI sounds, the easier it will be for consumers to see it fitting into their lives.

 

It also makes AI even more formidable in both personal and business settings if the AI can be both higher IQ and EQ than most of us pesky ol’ humans.

 

Businesses are catching on: More public cos are citing AI as a “friend” or “companion” on earnings calls in an effort to make B2C services more personal.

 

Mercedes-Benz, for instance, envisions drivers chatting with its MBUX voice assistant (which is built on ChatGPT) like a friend.

mercedes-benz sociability

Source: CB Insights — Mercedes-Benz Q4’24 earnings call

Beyond emotional connection, AI developers are also making their chatbots more useful through agentic capabilities. This dual focus on both feeling and function will shape the competitive landscape moving forward.

 

While still constrained by reliability and access issues, web-browsing agents like OpenAI’s Operator signal a future where humans interact regularly with AI agents that take actions on their behalf.


Already nearly every tech giant, plus leading LLM developers like OpenAI and Anthropic, have developed AI agents or are building out tools for others to develop them.

ai agent arms race

In a head-to-head matchup, OpenAI and Meta would bring distinct sets of advantages.

 

OpenAI has already built user habits with ChatGPT and is now layering agent capabilities on top. Meta must convince users its AI offering brings something meaningfully different.

 

But Meta’s experience in the social realm, combined with its existing algorithms, could give it an edge in creating AI that feels like it belongs in human conversations.

 

Watch for both companies to rapidly iterate on their consumer agent offerings in the coming months.


Customers can read up on 4 AI agent trends we’re watching this year.

TLDR — Tech loves drama, right?

Here's a roundup of recent tech drama:

  • Opti-no: Y Combinator deleted a demo from one of its YC W25 cohort companies, Optifye, from its social channels. The clip, which highlighted the company’s “Al performance monitoring for factory workers,” drew backlash from X users calling it “sweatshop” software.
     
  • Booz cruise: The Trump administration is reviewing $65B in government contracts across 10 consulting firms. Booz Allen, which generates 98% of its revenue from federal work, is one of the most exposed. CEO Horacio Rozanski says the firm is aligned with the administration’s priorities, but uncertainty looms as agencies must now justify why these contracts are “mission critical."
     
  • Into the fire: Mozilla drew heat over its Firefox web browser’s new Terms of Use, which included a broad statement that appeared to grant it rights to user data for AI use. Rival browser maker Brave Software’s CEO Brendan Eich posted “W T F” in reaction on X. Mozilla clarified that it does not plan to “use people’s data for AI or sell it to advertisers."
brendan eich post X

Source: X

  • Bad UI: Citigroup accidentally credited a client account with $81T instead of the intended $280 in an incident last April that recently came to light. The error came about in part thanks to clunky software, as Financial Times reports: “the amount field came pre-populated with 15 zeros, which the person inputting a transaction needed to delete, something that did not happen."

    I love you.

     

    Anand

    @asanwal 

    Co-Founder & Exec Chair

     

    P.S. On Thursday, you can hear directly from our analysts on the AI agent opportunity in commerce. Before then, catch up with our free report on the future of the customer journey.

    Get started with CB Insights

    Start your free trial

    CB Insights' emerging technology insights platform provides all the

    analysis and data from this newsletter. Our data is the easiest way to discover and respond to emerging tech. 

    Was this email forwarded to you? Sign up here

    X
    LinkedIn
    CB-Insights-Icon-Light

    Copyright © 2024 CB Insights, All rights reserved.

    498 7th Avenue, NY, CB Insights, New York,10018

    About Us | Update Preferences | Research | Newsletter