I Built a Recipe App by Texting My AI Assistant

I wanted a recipe app that actually understood cooking. Not just a database of recipes, but something that got the process. So I opened Telegram and started texting my AI assistant, Squidworth.

The research phase was the fun part. We analyzed five competitors—Paprika, Yummly, Tasty, Allrecipes, ChefGPT. Found five major features nobody had: pantry management, timeline view for cooking steps, step-by-step hands-free mode, leftover planning, and voice cooking assistance. We named our personas "Overwhelmed Oliver" and "Busy Brianna." It felt like having a product manager who never slept.

Then we started building. Through Telegram. I'd send a message: "Set up the Next.js project with TypeScript" and get back shell commands. "Create the database schema for ingredients" and get SQL. It was pair programming via chat, asynchronously, whenever I had a spare moment.

The deployment was where reality hit. Clerk middleware broke on Vercel's Edge Runtime. Prisma couldn't connect to Supabase through their IPv6 pooler. Foreign key constraints failed because my dev user didn't exist. Every fix uncovered another issue. I started keeping a running list in our chat, and Squidworth would check off items as we solved them.

Three days of debugging later, it worked. I pasted an AllRecipes URL, hit import, and watched "Spinach and Feta Turkey Burgers" appear in my database with properly parsed ingredients and steps. Live at cook3k-n959.vercel.app. Built entirely through conversation.

The lesson? The barrier to building isn't technical knowledge anymore. It's having someone (or something) to think through problems with. Turns out an AI assistant in your pocket works surprisingly well for that.

Next: Why I named it "Galactic Culinary Codex" and what that says about my branding judgment.