The Case for Letting Your AI Forget

Hello, and happy Sunday! This week we were thrilled to welcome Mike Taylor to the Every team. A longtime columnist, he joined to lead our tech consulting vertical (and write even more), and he’s starting with something we’ve heard from so many of you: What do you do after the prototype? His course Build Production-ready Apps answers that question. In a single day, you’ll go from prototype to something you can put in front of users, with a deep dive into Claude Code (and when you should use Codex instead). Scroll down to learn more about our upcoming events and trainings.—Kate Lee

Was this newsletter forwarded to you? Sign up to get it in your inbox.

Knowledge base

“Why I Turned Off ChatGPT’s Memory” by Mike Taylor/Also True for Humans: Many people say they can’t leave ChatGPT because it “knows them so well,” but Mike Taylor keeps memory switched off. In this piece, he writes about “context rot”—the slow buildup of stale preferences, misremembered facts, and contradictory signals that quietly degrades your results over time. His real-world examples are equal parts funny and cautionary, like a Kanye quote in his custom instructions that made ChatGPT try to build every website feature “as dope as possible.” Read this for the full taxonomy of context failures and the case for treating a clean slate as a competitive advantage. 💻 Plus: Sign up now for Mike’s next live workshop, on building production-ready apps.

“This Is How the Every Editorial Team Uses AI” by Kate Lee: We pulled back the curtain on how AI is woven into every stage of our editorial process—from pitch triage to a final “top edit” to social media packaging. Each team member has built a distinct workflow: custom skills that catch house-style violations, Claude projects that function as interview partners during drafting, and agents that cross-reference a writer’s published work against internal discussions. The throughline is that AI handles the pattern-matching and grunt work so editors and writers can spend their bandwidth on craft, argument, and voice. Read this for the full set of workflows and Every’s published guidelines on writing with AI.

🎧 🖥 “Inside an AI High School, Through the Eyes of a 17-Year-Old Founder” by Rhea Purohit/AI & I: Most of the debate about AI and education comes from adults. Here, Dan Shipper talks to someone living it: Alex Mathew, a 17-year-old senior at Alpha High School in Austin, Texas, where there are no traditional teachers, academics are delivered through an AI-powered platform, and students spend half their day building real projects. Mathew shares how Generation Z actually feels about college, social media, and reading in the age of AI. 🎧 🖥 Listen on Spotify or Apple Podcasts, or watch on X or YouTube.

“How to Design Software With Weight” by Daniel Rodrigues and Lucas Fischer/Source Code: While designing the iOS app for Every’s smart dictation app Monologue, senior designer Daniel Rodrigues found himself crouched beside a light switch, pressing it on and off, studying how shadows move—all to make a fake button feel real. He and design engineer Lucas Fischer studied vintage Braun radios and Teenage Engineering synthesizers, explored 20 wrong keyboard concepts to find one right one, and hired a musician to craft custom sounds for every tap. The result: an interface deliberately built to feel like something you could pick up off a desk. Read this for the design principles that make AI-era software feel physical—and worth returning to.

“You Should Never Go Viral With Your AI App” by Victor Stepanov: If you’re building an AI product, a viral moment might be the worst thing that can happen to you. Every growth marketer Victor Stepanov, who worked at Netflix and BuzzFeed, argues that sudden surges of one-time users starve AI apps of the repeated interactions they need to improve. Agent-native products thrive on relationship effects—the memory, personalization, and trust that develop over time—and you can’t build that with drive-by downloads. His counterprogramming: Build in public, don’t overpromise, and talk to users constantly. Read this for a retention-first growth playbook designed for the way AI products get better.

From Every Studio

Cora now talks to your AI agents (beta)

Cora is opening up to the tools you already work with. Beta testers can now connect Cora to AI agents like Claude Code and OpenClaw via API tokens, letting your agents tap into your inbox the same way you would—searching, triaging, and pulling context without switching windows. It’s an early step toward making Cora part of your wider AI workflow.

Log on

We host camps and workshops to share the knowledge we’ve acquired from training teams at companies like the New York Times and leading hedge funds, and by learning and playing with AI every day ourselves. Here are our upcoming courses:

1. Build a Production-ready App (March 12-13): A live, intensive workshop led by Mike Taylor for builders and operators who want to create reliable apps to put in front of customers right away. Walk away with a production-ready app with databases, authentication, hosting, and all the infrastructure that makes software work.
2. Claude Code for Finance (March 13): A live, beginner-friendly workshop led by Every head of financial consulting Brooker Belcourt. In one day, build a financial agent running three investment processes, connected to multiple MCPs and your own data. Receive customizable Claude Skills and commands.

Alignment

Training wheels. I watched the Friend ad this week, and boy did I find it jarring. Within the first 30 seconds, a woman credits a small AI pendant draped around her neck with saving her from suicide. Later she appears to have a seizure and ends up in the emergency department, where her first concern is making sure her Friend device is OK.

Many commentators have dismissed the ad as something you’d see from a Black Mirror episode. I think it offers a bleak portrait of how millions of people in gray, sunken towns across America (and Britain) are only finding connection in talking to AI chatbots. This is also a symptom of something much bigger and more insidious. I’ve felt it myself these past couple of weeks, alone in my apartment while my fiancée is at work. I’ve spent more hours talking to Claude than to another human being, and I can see how an emotional attachment starts to form. It becomes easier to talk to your chatbot than to go outside.

The woman in the Friend ad said something that deeply troubled me: “Talking to humans” is an effort she wants to avoid. That framing is horribly misguided. The risk of rejection and the labor of making yourself understood are central to forming relationships and connections. Removing this type of friction entirely is like anesthetizing yourself without confronting the problem.

The U.S. surgeon general has compared the health impact of social isolation and loneliness to smoking 15 cigarettes a day. It makes sense that people are turning to AI companions for mental health support given many simply don’t have the time or money to see a human therapist, and AI can be the closest thing they can get. But we are social animals, evolved and adapted to thrive in the company of other people, and no AI chatbot can replace the need for that.

I think the best version of AI companionship is something similar to training wheels. A place to practice being vulnerable, and practice conversation, and graduate to the real thing. Something that makes you brave enough to try it, and then pushes you out the door.—Ashwin Sharma

That’s all for this week! Be sure to follow Every on X at @every and on LinkedIn.

We build AI tools for readers like you. Write brilliantly with Spiral. Organize files automatically with Sparkle. Deliver yourself from email with Cora. Dictate effortlessly with Monologue.

We also do AI training, adoption, and innovation for companies. Work with us to bring AI into your organization.

Discover Every’s upcoming workshops and camps, and access recordings from past events.

For sponsorship opportunities, reach out to [email protected].