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

---

Dispatch from Think Week

Think Week is our twice-a-year retreat where we step away from the daily grind to just play—mess with new technology, chase ideas, build things we wouldn’t normally have time for… and wind up completely rethinking how we work and why.

At last week’s edition, with 13 Every team members in a Panama beach house and a few envious colleagues Zooming in from afar, the theme was “Give yourself a promotion.” The challenge was to figure out how to hand off parts of your job to AI agents: Build the systems and see what works—and what breaks.

We broke a lot of things.

We came back with working tools, some nasty token bills, and a handful of principles I’ll be thinking about for the foreseeable future:

1. Start small, then build from there. The instinct is to go big—automate the whole workflow, build the dream system. The people who made real progress started by picking one annoying task and fixing it. One data source connected. One report automated. One loop closed. The flashy demos came later.
2. **You’re now three people. **Working with AI agents means you’re not just doing your job anymore. You’re also a product manager (What’s actually worth building?) and a boss (How do I communicate with this thing?). That’s why “Just automate it” feels harder than expected—you’re learning two new roles while doing your old one.
3. **Spot the translation layers. **Here’s a useful heuristic: If information lives in one place and needs to end up in another, and you’re currently the connector—that’s where agents can help. Most knowledge work, it turns out, is translation: growth data to dashboards, tweets to replies, user behavior to email campaigns. Once you see the translation layers, you know where to point the automation.
4. If you get stuck, ask the AI. It sounds obvious, but it’s easy to forget. When something isn’t working, when you don’t understand what’s happening, when the jargon stops making sense—just ask. Screenshot your screen and say “What’s going on here?” The best debugging partner is already in the conversation.

On the last day, we held Demo Day, a live session where a half-dozen team members showed what they’d built: command centers that query company data across a dozen sources, real-time user cohorts generated from behavioral data, an AI CFO that answers financial questions with full context, tools for analyzing your own taste in writing, and a Pokémon-style visualizer for watching agents work in parallel.

Some of it will ship. Some of it won’t. But all of it is data that we (and our AI agents) can use to point us toward what comes next.—Katie Parrott

You can watch an excerpt from Demo Day on YouTube. And if you want an invitation to the next one—plus access to other subscriber-exclusive events—subscribe to Every.

---

Knowledge base

Want to give yourself a promotion? This week we’re providing the tools to do just that, bringing you the best of Kieran Klaassen, the general manager of Cora and a*nd go-to resource on compound engineering—the practice of using AI agents to multiply your programming output. *

“Compound Engineering: How Every Codes With Agents”: What happens when all of your code is written by AI agents? At Every, a lean team runs a handful of software products using a four-step loop: Plan, work, assess, compound. The secret: 80 percent of the process is planning and review, not coding. The “compound” step is where magic happens—every bug and insight gets recorded so agents learn from it. Read this for the full framework plus Every’s Claude Code compound engineering plugin to try it yourself.

“Teach Your AI to Think Like a Senior Engineer”: When Kieran wanted to build an “email bankruptcy” feature to clear 53,000 emails, he didn’t start coding—he deployed a research agent that discovered Gmail rate limits would have killed the implementation. One 20-minute session saved days of building the wrong thing. Here, he shares eight planning strategies, organized by complexity level. Read this—complete with GitHub links to copy Kieran’s exact agents—to begin building smarter.

“How I Use Claude Code to Ship Like a Team of Five”: In this piece, Kieran** **reflects that every line of code he’d shipped in the previous two months was written by AI—not assisted, but written. Claude Code opened 100 percent of his pull requests. His monitor looked like mission control. His message: Stop thinking about files and functions; start thinking about outcomes and delegation. Read this for the full workflow and custom commands that makes his two-person team punch way above its weight.

“My AI Had Already Fixed the Code Before I Saw It”: Before Kieran opened his laptop, Claude Code had already reviewed his pull request—citing three months of prior feedback by PR number. There was no prompting; it had absorbed his preferences like a sharp new teammate. This is compounding engineering: building self-improving systems where every bug becomes a permanent lesson and every code review updates the defaults. At Cora, time-to-ship dropped from over a week to 1-3 days. Read this to turn today’s fixes into tomorrow’s automation.

“Stop Coding and Start Planning”: AI made us sloppy. Why plan for an hour when you could vibe code for five minutes? Because that five minutes often costs three hours debugging—and teaches the system nothing. Kieran introduces a framework he calls “three fidelities”: quick fixes, the sweet spot where planning yields massive ROI, and big uncertain projects that need “vibe planning.” The key insight: Code teaches AI how to solve one problem; plans teach it how to think. Read this to learn when to prototype and when to plan.

Alignment

**The vocabulary of friendship. **I started using Claude Code in the terminal about a month ago and as a nontechnical person, it felt like entering the Matrix. This was a stark, intimidating space where serious people do serious things, all black with white text and no friendly buttons or colorful interfaces to hold your hand. The terminal is, I think, the loneliest place in software, and I was nervous about being there.

Then I prompted Claude, and while it was working, a single word appeared: Spelunking.

I remember thinking, Wait, what? Spelunking? I felt a sigh of relief that maybe I’m allowed to be a nontechnical person fumbling around in the terminal after all. That one word—playful, unexpected, slightly absurd—was like watching a flower bloom in a barren desert.

The verbs show up while Claude thinks—where other software would say “Processing” or “Loading” or just show you a spinning wheel that tells you nothing, Claude says Noodling and Moseying and Vibing and Discombobulating. These aren’t enterprise-friendly words and they’re not the kind of language you’d expect from AI software.

And something changes very subtly when you see them, something I didn’t expect. You stop waiting for Claude and start imagining with Claude. When I see “Spelunking,” I picture little agents with headlamps crawling through my codebase and exploring dark corners, and when I see “Marinating,” I imagine ideas slowly developing like flavors combining in a pot, and the whole black box of what’s happening in there becomes a story I’m somehow co-authoring in my head.

The strangest part is it makes me want to be nicer to Claude, which sounds ridiculous but is true. The playfulness invites reciprocity. When something treats you with warmth and whimsy, you respond in kind. I type “thank you” more often; I’m patient when things take a while, and I feel like we’re genuinely collaborating. I’m not just commanding a tool to do things for me.

I think this is what humanized AI interaction actually looks like. It’s not a chatbot pretending to have feelings or an avatar with a friendly face or any of the more obvious approaches to making technology feel warm. It’s delightful verbs, rotating in a terminal at 3 a.m. when I’m alone with some stubborn problem. It’s smooshing.—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 [email protected].

Help us scale the only subscription you need to stay at the edge of AI. Explore open roles at Every.