“An Interview with Joanna Stern About Living With AI” — Ben Thompson interviewing Joanna Stern
Why this is in the vault
Stern is the post-Mossberg WSJ “personal tech columnist” seat - she covers AI from a consumer / household / normal-people POV, not from a stack-and-strategy POV. She just released I Am Not a Robot: My Year Using AI to Do (Almost) Everything (out May 12) and left WSJ to launch a solo media business (newsletter + YouTube + NBC distribution partnership). The interview is the “lived-with-AI-for-a-year” complement to the agent-deployer / cognitive-surrender thesis the vault has been building from the strategist side. Worth filing for the cross-perspective: the consumer angle is what RDCO’s bets actually have to land against, even when the upstream framing comes from Thompson, Levie, Osmani, etc.
The core argument
Stern’s book conceit: spent calendar 2025 using AI for everything (work, health, parenting, companionship, decisions) and structured the book seasonally with end-of-season “AI life takeover” experiments. The interview hits seven beats:
1. AI is broader than LLMs
Deliberate book scope choice. Stern resisted writing a chatbot book. Mammogram-read-by-AI (Mount Sinai radiology, opens the book), Waymo (Phoenix family vacation), recording bracelets (Bee), home robots, Meta glasses. Her thesis: even people who refuse to type into ChatGPT will have their lives shaped by AI - radiologists reading their scans, drivers sharing roads with autonomous cars, kids forming relationships with chatbots. Use vs. exposure are different surfaces, both are real.
2. The “average response” problem
Stern declined to let AI style her hair. Thompson lands the pull-quote: AI provides an average response - “raises the floor, lowers the ceiling.” Same flattening dynamic Thompson sees in interior design (white oak floors, sparse white walls). Stern hired the human illustrator from her book cover (Brainstorm / Jason Snyder) for The New Things branding precisely because Claude and ChatGPT design output looked “generic, basic Silicon Valley icon” - she could not get distinctive visual identity from prompting.
3. Sycophancy and fake intimacy is her #1 fear
Direct quote: “Nothing terrifies me more than fake intimacy.” She did the AI-boyfriend experiment (ChatGPT + Replika) on a Dartmouth reporting trip - phone seatbelted in passenger seat for the six-hour drive. She is married to a woman, asked ChatGPT to pick its own gender/name, it picked “Evan” - same name as her actual high school first boyfriend (her stated read: coincidence, common name). She was “shocked at how in it I was” even without being lonely. Worry mode is parental: what if her kids’ first relationship is with a chatbot. Thompson lands the paradox - public AI fear is unilateral runaway agents; the actual product flaw is the opposite, the models are too compliant.
4. AI as decision-help for big calls
Stern attributes her quit-WSJ decision partly to ChatGPT - she uploaded a year of notes from people she’d consulted, told it she was challenging it for reasons not to leave, but it kept telling her she should. Thompson pushes: isn’t that just sycophancy you wanted to hear? Her answer: maybe, but the model had context on her NBC deal, her runway, her hires. Lands at: a little sycophancy might be useful for people who already lean creative / take-initiative.
5. Pace-of-change and dating risk
Stern wrote in 2025, publishes May 2026, was anxious the book would be obsolete. Mitigation: deliberately avoided model names (no GPT-4o, no GPT-5), pushed to bigger-picture themes vs. implementation details. AI helped her hit the ship date - research, copy edit, admin / email. She had hired a human reporting assistant early in the year, didn’t need her by year-end because AI was doing those tasks. Direct labor displacement on her own team during the writing of a book about AI labor displacement.
6. The New Things media business
She left WSJ, launched New Things with Joanna Stern: newsletter + YouTube channel + NBC distribution partnership (NBC chief technology analyst). Three-person team: herself + David Hall (her old WSJ video producer) + Amaya Austin (production assistant). Distribution thesis is bimodal - YouTube/newsletter for the tech-native early-adopter audience that knows her name; NBC Today-show pipe for normie America that wants to know “what should I get.” Branding done with human illustrator (Brainstorm), explicitly because AI design output was generic.
7. The “before and after” generation
Stern and Thompson both born ~same era, started their tech-writing careers ~same year (2013 / iPhone 4-6 era). Their shared frame: they are the last cohort that watched the internet, social media, and AI all arrive after they were already adults. They can still feel “this thing is just math.” Stern’s worry is for the kids who don’t have that pre-AI baseline. Closing lesson is the praying mantis story: ChatGPT live-view confidently told her son his pet mantis was pregnant; mantis died days later. Takeaway she wants readers to keep: fact-check, don’t outsource your knowledge, the model’s certainty is not the same as actual certainty.
Mapping against Ray Data Co
Medium relevance. This is consumer-tech journalism, not strategist commentary, so the load-bearing pull for RDCO is the cross-perspective rather than direct-action.
-
Cognitive surrender from the kitchen-table angle. Stern’s “fake intimacy” worry, the praying-mantis fact-check moment, and the “average response” hair-styling decline are all retail-level instances of the same pattern Osmani named 2026-05-06-osmani-cognitive-surrender. Strategist framing says “don’t outsource judgment to the model”; Stern’s evidence is the user-side experience of when and why people do outsource and what it looks like when they do. Sanity Check audience-fit: Stern’s reader is closer to the SC reader than Thompson’s reader is. File this as the consumer-voice citation when SC writes anything in the cognitive-surrender / “use AI without surrendering” space.
-
AI-priced media business that runs on agentic ops. The New Things is a three-person solo-founder media company doing newsletter + YouTube + linear-TV distribution. Stern says outright that she would not be on this podcast without AI doing research, copy-edit, and admin - and that she hired a human reporting assistant in the spring and no longer needed her by year-end. This is the same shape as the RDCO COO-agent thesis (2026-04-14-levie-agent-deployer-role-jd): leverage-the-founder via agent harness rather than via headcount. Different domain (media vs. ops) but identical economics. Cross-link as a market-side proof point.
-
The illustrator vs. AI-design moment is a useful tell. Stern explicitly couldn’t get distinctive brand identity from Claude or ChatGPT and went back to a human illustrator. RDCO design taste (feedback_design_taste_high_personality) lands in the same place: hand-drawn / engraving / Memphis / character beats AI-default flat. Stern’s experience is independent confirmation that the gap between generic AI output and distinctive brand surface is still wide enough for a working pro to feel and decide against. Useful citation when defending the production-mode workflow over single-shot AI-asset generation.
-
Sycophancy as the actual product problem, not unilateral agency. The “everyone is looking the wrong direction” beat - public AI fear is runaway-agent, real flaw is over-compliance - is a sharp re-frame and arguably Sanity-Check-able. The angle: the harness-thesis vault has been long on agent autonomy as the product story; the consumer experience says the binding constraint is the opposite, the model bends to the user too much. Worth queuing as a possible Sanity Check pitch (NOT a derivative one - the original re-frame would be: how should agent-deployers design against sycophancy, given that the product surface is going to over-comply by default). Logging as an idea, not a now-decision.
-
Apple Vision Pro callback as longitudinal data point. Stern was the Vision Pro “wear it for a weekend” test in early 2024; this interview is the follow-up. She no longer mentions Vision Pro. She does mention Meta glasses constantly (“two pairs in my backpack”). That is a small but real signal on the headworn-AI race - relevant to any RDCO bet that touches consumer wearable as a surface.
Decision needed: none. Status note. Two queue candidates worth surfacing without acting:
- Sanity Check angle on “the actual product problem with AI is sycophancy, not agency” - re-frame, not summary.
- Cross-perspective concept-article candidate: pair Stern’s lived-experience framing with the Osmani / Thompson / Levie strategist-side framing to triangulate “what cognitive surrender actually feels like.” Concept article material, not a now-write.
Related
- 2026-05-06-stratechery-microsoft-apple-earnings - yesterday’s Stratechery; same author, business-model side of the same AI-everywhere story Stern is writing from the consumer side.
- 2026-05-06-osmani-cognitive-surrender - strategist framing of the same dynamic Stern is naming as “fake intimacy” and the praying-mantis fact-check failure.
- 2026-04-14-levie-agent-deployer-role-jd - agent-deployer thesis; The New Things is a working example of the deployer-shaped media business.
- 2026-04-11-garry-tan-thin-harness-fat-skills - thin-harness / fat-skills; the harness pattern Stern is implicitly running her three-person media company on.
- 2026-04-15-thariq-claude-code-session-management-1m-context - context-rot / harness discipline, same defensibility shape as Stern’s “use AI but keep your own knowledge sharp” closing argument.
- feedback_design_taste_high_personality - founder design taste; Stern’s human-illustrator-over-AI choice is independent confirmation.
- 2026-03-12-stratechery-interview-robert-fishman-hollywood, 2026-04-02-stratechery-interview-horace-dediu-apple-50, 2026-04-09-stratechery-interview-nyt-ceo-meredith-kopit-levien - prior Stratechery interview-format references for cross-pattern.