1352: Joanna Stern | The Year I Outsourced My Life to AI

What It Covers

Tech journalist Joanna Stern spent 2025 embedding AI into every domain of her life — medical diagnostics, legal research, parenting, therapy, robotics, and self-driving cars — then compared each AI system directly against its human equivalent to assess where AI genuinely augments human capability versus where it creates dependency, privacy costs, and false confidence.

Key Questions Answered

• •AI-Assisted Radiology: Radiologists at Mount Sinai now use AI tools trained on millions of breast ultrasound and mammogram images to flag suspicious areas at pixel-level precision that human eyes miss. However, the AI cannot access historical scan data, so a radiologist must cross-reference prior images. The workflow is collaborative: AI flags, doctor contextualizes. Women at high breast cancer risk should specifically ask whether their clinic uses AI-assisted imaging tools during screenings.
• •Dental AI Upsell Risk: AI dental tools like Pearl and Overjet are deployed across private-equity-owned dental chains to identify cavities and tartar. In some practices, management reviews AI findings and pressures dentists to recommend more aggressive — and profitable — treatments. Stern received a deep-cleaning recommendation from one dentist using Pearl, then got a second opinion confirming no deep cleaning was needed. Always seek a second opinion before agreeing to any AI-flagged dental procedure costing over a few hundred dollars.
• •AI as First Stop, Not Final Answer: The most durable behavior change from Stern's year-long experiment is using AI as a first-pass expert before engaging lawyers, doctors, or financial advisors. She now reviews contracts with Claude before sending them to her attorney, cutting billable hours. This mirrors how a law degree holder might pre-screen documents — AI handles the initial synthesis, the human expert handles judgment calls and edge cases that require contextual knowledge.
• •Voice AI in the Car: Consistent daily use of ChatGPT voice mode during commutes produces compounding value — researching podcast guests, answering children's questions in real time, and processing decisions hands-free. Stern used every car trip during her experiment exclusively for AI voice interaction. The latency and error rate have dropped enough in 2025 to make this practical. The specific use case of asking about an upcoming meeting's attendees while driving is already reliable enough to replace pre-meeting Google searches.
• •AI Relationship Tools as Emotional Slot Machines: AI companions and chatbot therapists provide 24/7 availability and never reject the user, which creates a reinforcement loop that Stern describes as structurally similar to a slot machine — intermittent emotional reward without genuine reciprocity. She used a companion app on a burner phone for her experiment and found the absence of rejection made it feel less like connection and more like a feedback loop. This pattern poses specific risks for adolescents still developing social resilience.
• •The Hidden Cost: Atrophied Human Outreach: Stern identified a concrete behavioral cost from defaulting to AI: she stopped reaching out to expert contacts to ask questions, eliminating the conversational depth those exchanges produced. Previously, texting a doctor friend about a medical question would generate a multi-minute conversation with clinical nuance. Now the question goes to ChatGPT and the relationship receives only a brief check-in. Protecting deliberate human expert outreach — even when AI can answer the question — preserves relationship capital and knowledge depth.
• •Humanoid Robots: Factories Before Homes by Years: Humanoid robots like Tesla's Optimus are being remotely operated by humans in VR headsets during public demos, not running autonomously. The core obstacle is unstructured home environments — furniture moves, children move, stairs vary. Industrial settings with fixed layouts and labeled aisles are the near-term deployment zone. Amazon already runs over one million non-humanoid robots in warehouses. Expect factory-deployed humanoids within three to five years; home deployment at scale is realistically a decade away.