why every frontier lab employee is being vague right now


Over the past few weeks, OpenAI and Anthropic employees have been posting vaguely about something.

Nobody would explain what it was or go into any detail. Just hints.

I think I know why. And once you understand it, you'll realize this is the most important thing you could be working on right now.

Here's what's going on.

Opus 4.8 leveled the playing field. Anyone can build anything now. Which means just using AI is no longer a competitive advantage.

So what is?

Loops.

An automated loop is a system that lets your AI work autonomously, continuously shipping and reviewing code, 24/7/365, while your competition can only build when they're sitting at their keyboard.

That's why nobody at the frontier labs is explaining this. Loops are true technical IP. If Anthropic shared their internal loop system with the world, their team would lose their edge overnight. So they're keeping it vague. On purpose.

Three weeks ago I thought loops were impossible. Now I think they're mandatory.

Here's the thing. If you and a competitor are building the same product, but you have a loop and they don't, you'll destroy them. It's not even close.

Here's how to get started today.

1. Use the most powerful model available

Loops require serious thinking. The smarter the model, the better the output. Don't cheap out here.

2. Create a skill called /spec

Have your agent build a skill that asks you enough questions about the feature you want to build until it fully understands it in detail.

3. Save the spec to a second brain

I use Linear. Whatever you use, make sure the spec lives somewhere your agent can reference it.

4. Make skills for /build and /review

Two skills. One builds. One reviews. Both operate from the specs that got created in step 3.

5. Run the loop

Now your agent has everything it needs. It specs, it builds, it reviews. On repeat. Without you sitting at the keyboard.

6. Let it compound

The more you run it, the better the skills get. This is not a one-time setup. It's a system that improves every time you use it.

The game pieces moved. The battlefield shifted. Just using AI is no longer an advantage.

Now you need to automate.

Want to build your first loop live? I run a session inside the Vibe Coding Academy every Friday. Come build with us.

Join the Vibe Coding Academy here.

Keep shipping,

Alex

Ship/It Weekly

The weekly newsletter for founders and builders using AI to ship faster—practical tactics, tools, and real-world lessons from Alex Finn.

Read more from Ship/It Weekly

AI has gotten expensive. Especially when it comes to API pricing. Last month alone, I burned $1,300 on Claude tokens inside Hermes. Luckily, Claude Sonnet 5 has flipped the math. It's almost as good as Opus 4.8, but for a fraction of the price. By far the best bang for your buck in AI right now. Does that mean Sonnet 5 completely replaces Opus? Not for me. But it's changing how I use AI. Which is why I put together a full breakdown of everything that changed with Sonnet 5, and exactly how I'm...

In 2025, I made $500,000 with Claude Code. You might assume I made it through hard work over many years to build one killer app. But that’s not the case. The app was actually the easy part. Claude Code ships a working V1 in minutes now. Anyone can do it. So the app can't be your edge. And it was actually never where the money lived. So where did the $500k actually come from? Mostly from Creator Buddy, a SaaS I built with Claude Code that does over $300,000 in ARR. But the other ~$100,000 came...

One-person companies are building million-dollar mobile apps using Claude Code. If you've been sitting on an app idea and waiting for the right moment to build it, this is it. I put together a full walkthrough of everything you need to know to build and ship your first iOS app, even if you've never written a line of code in your life. Here's what I cover in the video: How to set up Xcode and get Claude Code working in your project folder (two free tools, five minutes of setup, and you're...