The AI coding stack I actually use: what's in my workflow

A developer details their daily AI coding stack, using both an editor-integrated AI tool and a chat model for different tasks, and notes that AI excels at boilerplate and test scaffolding but is unreliable for security-sensitive code. They pay for two tools but advise learners to start with one paid tool plus free tiers, and emphasize maintaining a personal library of effective prompts.

The Primer · Tools & Stacks The AI coding stack I actually use: what's in my workflow No affiliate links · vendor-neutral · nothing here pays us No affiliate links in this post. Nothing here pays me. Just what I actually reach for after a year of building with AI tools daily, and where each one stops being worth it. For writing code: I bounce between an AI-in-the-editor tool and a chat model in a browser tab. The editor integration is faster for small, in-context edits; the chat window is better when I need to think through a problem out loud before touching code. Using both, not picking one, is the actual answer most “X vs Y” posts won’t give you. Where it shines Boilerplate, test scaffolding, unfamiliar APIs, and “explain this legacy file.” Genuine multi-hour-per-week savings, no exaggeration. Where it stops being worth it Anything where being subtly wrong is expensive: auth, payments, security-sensitive logic. I still write those by hand and use AI only to review. The tool is a strong junior, not a senior; I don’t merge what I can’t read. The subscription math I pay for two tools. It’s real money each month, and it’s worth it for me because I ship for a living. If you’re learning or building nights-and-weekends, one paid tool plus free tiers is plenty, don’t let anyone guilt you into the full stack before you’re earning from it. The one habit that matters more than tool choice Keep a running file of prompts that worked. Your own library beats any tool’s defaults within a month. What’s in your setup, and where have you found the AI tools stop earning their keep?