Hey there,

In the vibe-coding world, "plan before you build" is having a moment. There's a whole movement around it, and people love saying that building with AI is something like 95% planning, 5% actually writing code.

Dharmesh Shah recently wrote about the shift from "prompting" to "creating loops" to put a framework to this. A loop has three parts: the objective (what you're trying to reach), the metric (how the system itself knows it's getting warmer or colder), and the boundary (how far it can run before it checks back in with you). Get those three right and you stop hand-cranking every prompt; you start designing the thing that cranks them for you.

My friend Kyle James has put that into practice particularly well, so I chatted with him about it on the newest installment of the AgentGTM podcast (with a live demo and open-sourcing of the skills he uses).

Here are some things we cover:

• The "measure twice, cut once" planning method Kyle uses to build fast without breaking things
• A live, from-scratch build of a competitive-intelligence tool — narrated step by step
• The two-model setup that turns AI review into an adversarial back-and-forth until the plan actually holds up

Here are the skills that Kyle uses in our interview.

And links to listen:

• ​Spotify​
• ​Apple Podcasts​
• ​Youtube​

Thanks,

Shaalin

P.S. if you're early in your building journey, and want some help building AI tools, reply! I'm helping a few people build their own agents.