The Droid You’re Looking For

Hello, and happy Sunday! Was this newsletter forwarded to you? Sign up to get it in your inbox.

Knowledge base

“Inside the AI Workflows of Every’s Six Engineers” by Rhea Purohit/Source Code: Six engineers. Four AI products. A daily newsletter read by 100,000 people. How do we do it? Rhea Purohit went inside the workflow of each engineer at Every to find out—from Yash Poojary running Claude Code and Codex side-by-side on two separate machines to see which “personality” ships faster, to Nityesh Agarwal doing everything on a single MacBook Air with his finger hovering over the Escape key like a nervous parent. Read this if you want to see how engineers structure their days and what tools survived the hype cycle.

“How to Use Claude Code Like the People Who Built It” by Rhea Purohit/AI & I: Cat Wu and Boris Cherny, the founding engineers of Anthropic’s Claude Code, reveal how they use their own product—from making subagents fight each other to catch bugs, to setting up stop hooks that won’t let Claude quit until tests pass, to turning past code into reusable leverage. Listen to this interview with Dan Shipper if you want insider tactics for getting more out of Claude Code, straight from the people who watch hundreds of engineers use it every day. 🎧 🖥 Listen on Spotify or Apple Podcasts, or watch on X or YouTube.

“Vibe Check: Cursor 2.0 and Composer 1 Alpha” by Rhea Purohit/Vibe Check: Cursor dropped a refreshed code editor and its own blazingly fast LLM, and the Every team spent a few days stress-testing both. Alpha is genuinely fast and fun, though it falls short of Sonnet 4.5 on precision and UI generation. Read this for an honest assessment of whether Cursor’s comeback is strong enough to pull developers away from the terminal.

“Vibe Check: I Canceled Two AI Max Plans for Factory’s Coding Agent Droid” by Danny Aziz/Source Code: Danny Aziz was debugging a failing database migration in the back of a moving car when he switched from Claude Code to Droid and watched it solve the problem in one shot. Somehow Droid makes each model work better than it does in its native wrapper. Read this if you’re tired of losing context every time you need a different model’s strengths, or if you want to see what convinced a senior engineer to cancel his Claude and ChatGPT Max plans.

“How Tools Shape How We See the World” by Dan Shipper/Chain of Thought: Dan Shipper spent years searching for the perfect note-taking system—until he realized he was trying to become a machine. But language models revealed something he couldn’t see before: Our tools don’t just help us work, they determine what parts of reality become visible to us. Read this—the fifth and final piece in a series exploring AI’s new worldview—if you want to understand why the complexity of our tools literally changes the complexity of our reality.

“How My Company Turned a Data Crisis Into an AI-native Rebirth” by Stella Garber: When Slack cut off a critical data source with just four weeks’ notice, Hoop faced extinction. Instead of panicking, founder Stella Garber and her team used the crisis to fundamentally rebuild—not just their product, but how they build products. And they did it with AI. Read this if you’re wondering what it looks like when you’re leading a team and the ground shifts beneath you.

Coming up: Claude Code for Beginners

On November 19, Every CEO Dan Shipper will be hosting a workshop that’s perfect for anyone who’s been wanting to learn Claude Code. The day-long session will take you through installation and setup to a shipped project and a repeatable, reusable workflow—no coding experience required. Learn more and sign up.

Alignment

The 10 percent rule. Last week it was Novo Nordisk partnering with Anthropic to cut clinical documentation from 10 weeks to 10 minutes. This week Eli Lilly announced the world’s largest pharma-owned AI factory. It has more than 1,000 Nvidia Blackwell Ultra GPUs doing 9 quintillion calculations per second. The pharma company’s goal is to let AI agents suggest molecular combinations that human chemists would never have time to explore.

What makes this particularly interesting is that roughly 10 percent of drug candidates make it through phase 3 trials to pharmacy shelves. The other 90 percent fail because the drugs are too toxic, not effective enough, or have side effects that only surface after months of continued use.

This makes drug discovery the rare domain where AI hallucinations aren’t a bad thing. So what if an AI spams 10,000 suggestions, some of which are nonsense? They’re easily discarded, and at a fraction of the cost (which is pretty much all compute) of synthesizing and testing one molecule (which takes months and runs millions of dollars). Eli Lilly isn’t using raw AI output, either. It’s training foundation models on $1 billion of proprietary data—decades of documented failures, every molecule that didn’t work and why. A chemist can dismiss a bad AI suggestion in seconds while synthesizing it takes weeks, so even if the AI generates more noise than signal, the cost structure still favors generating more options. AI’s first real win might be in being allowed to fail faster and more cheaply than we ever could.—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.

Get paid for sharing Every with your friends. Join our referral program.

For sponsorship opportunities, reach out to sponsorships@every.to.