How much should you tell your AI agent?
Oct 6, 2026 · 41m
Summary
David Pierce and Allison Johnson compare agentic AI messaging tools like Meta’s Muse, OpenAI’s Dots, and Instinct, analyzing their distinct approaches to automating daily tasks. The hosts discuss the trade-offs between convenience and privacy, highlighting concerns about data access and the potential for AI to make unintended purchases. They also explore the philosophical tension between human interaction and automation, using personal anecdotes to illustrate how these assistants fit into modern life.
Topics discussed
Welcome to The VergeCast and intro to AI assistant segment
Preview of AI assistant testing with Allison Johnson
90 Seconds: TVLand, Warner Bros, and Fitbit Edge news
90 Seconds wrap-up and Prime Big Deal Days ad
Sponsor: Charles Schwab On Investing
Personal motivation for testing AI assistants
The backlog problem and the need for simple automation
Comparing Meta Muse, OpenAI Dots, and Instinct
Which interface feels right: app, messaging, or AI chat?
Instinct's strengths: simplicity and iMessage integration
Dots as work software vs. Muse as a messaging app
The utility of constrained interfaces like Instinct
Browser limitations vs. Muse's connector ecosystem
Muse's desktop experience and visual feedback
The 'behind the curtain' feel of AI software
Surprising successes: the 'hero experience' of AI
The danger and convenience of automated transactions
Using Dots for work tasks and website updates
Iterative design: talking to AI for 10 minutes
The difficulty of adopting new AI behaviors
Moral feelings about AI and the desire for convenience
What it means to be a person in an automated world
Feeling 'icky' about AI usage and testing ethics
Convenience vs. human interaction: The Atlantic article
The loss of small human interactions (e.g., baristas)
Automating the last small tasks of the day
The value of manual tasks like doing dishes
The danger of accelerating task accumulation
The tension of AI: unwanted but assumed by others
Why low-friction tools like Muse and Instinct are useful
AI as a tool in a mixed toolkit, not a single assistant
Sponsor: Framer
Sponsor: Shopify
Sponsor: Shopify (continued)
Sponsor: Satellite Collar Go
Sponsor: Satellite Collar Go (continued)
Sponsor: Prime Big Deal Days
Lines in the sand: tasks you shouldn't automate
The fear of cognitive atrophy and reliance on AI
Catching yourself relying on AI for simple tasks
Using Muse to search for a specific car
The moment Muse almost bought a car
The alarming potential of AI with credit card access
The 'everything connector' problem and data privacy
Trust issues with Meta and data stewardship
The risk of password manager compromise
Advice for friends asking about Muse
The moat of two-factor authentication and security
Monitoring credit cards for AI-driven charges
Meta's thirst for engagement and sales
Using Instinct for daily email highlights
Automating professional development searches
The utility of aggregating information from multiple accounts
Verge Awards: Best AI Gadget is a smartphone
Call to action: Vote for Verge Awards
Outro and subscription reminder
Listen ad-free on Castria