{"id":110721,"date":"2026-06-22T12:00:00","date_gmt":"2026-06-22T12:00:00","guid":{"rendered":"https:\/\/www.red-gate.com\/simple-talk\/?p=110721"},"modified":"2026-06-15T09:02:22","modified_gmt":"2026-06-15T09:02:22","slug":"i-built-an-application-by-talking-to-ai-heres-what-happened-and-what-i-learned-part-one","status":"publish","type":"post","link":"https:\/\/www.red-gate.com\/simple-talk\/ai\/i-built-an-application-by-talking-to-ai-heres-what-happened-and-what-i-learned-part-one\/","title":{"rendered":"How I built an app just by talking to an AI LLM – the good, the bad, and what I’d do differently next time (part one)"},"content":{"rendered":"\n

Like most of you, I’ve been using AI large language models (LLMs)<\/a> at work and at hobby level for a couple of years now. Snippets of code, the odd data conversion, a small database for a presentation.<\/strong><\/p>\n\n\n\n

What I hadn’t done \u2014 and what kept nagging at me \u2014 was build an entire application, end-to-end, through AI. Nothing more, nothing less. So that’s what this series<\/a> is about. I’m building a real, small, useful application called dbRosetta, and I’m doing as much of it as possible by talking to an LLM.<\/strong><\/p>\n\n\n\n

The premise is deliberately tiny. Different databases give different names to similar concepts. A SQL Server stored procedure<\/a> isn’t quite a PostgreSQL<\/a> function, but the family resemblance is obvious. dbRosetta is a Rosetta Stone<\/a> for that. Pick a source database, pick a target, type a term, get the equivalent \u2014 plus a link to the docs and a short description. That’s it. Nothing clever.<\/p>\n\n\n\n

So, the app is not the interesting part – it’s how the app actually gets built. Let’s get into it.<\/p>\n\n\n\n

Before we begin: establishing the rules<\/h2>\n\n\n\n

Throughout this series, I’ll post the prompts and the responses. Anywhere the AI wrote text or code, I’ll mark it. The repo lives on my public GitHub<\/a>, under Apache-2.0. The app will run at scarydba.com<\/a> \u2014 I have the most control there, and if I do something silly I won’t be breaking Redgate<\/a>. The database is PostgreSQL on Azure Flexible Server<\/a>.<\/p>\n\n\n\n

For the LLM, I’m using Microsoft Copilot<\/a> (work pays for it, which is reason enough), with GitHub CoPilot<\/a> inside vsCode<\/a> for code generation, completion and validation. Whenever it hallucinates<\/a> or wanders off down a rabbit hole, I’ll step in. Otherwise, the rule is: let the AI drive as much as it sensibly can. Here’s my first prompt, and the response it gave me:<\/p>\n\n\n\n

Prompt<\/strong><\/td>Response<\/strong><\/td><\/tr>
I’ll be providing design specs and more, but initially, I’d like you to introduce yourself and your role in the project. Keep it to about three sentences since it’ll be part of a larger article.<\/em><\/td>I am Copilot, an AI development partner here to help design, build, and document dbRosetta. I\u2019ll assist with architecture and schema design, generate production-ready SQL and application code, and craft reusable prompts and CI\/CD workflows. I\u2019ll also review drafts, refine session abstracts, and help translate your design specs into tested, deployable artifacts.<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

So, the current tool list is:<\/h4>\n\n\n
\n