By skyforbes I am using AI Studio (w/ Gemini) at default settings, rarely using Grounding with Google or any "checkbox on the right" – with all to default i.e temperature at 1.0 and always latest Pro model.
I send a prompt, get unsatisfactory answer (either code doesn't work or I can see the LLM misinterpreted part of my request), I delete the answer and edit the prompt to submit it again. i rarely chain multiple chats.
How do YOU do it? what do you recommend me do? I do not like the "automatic vibe coding tools that just change stuff on a whim" 🙂 so I prefer to copy/paste myself, alternatively scan through the code / do little review, check for errors and think twice about what goes into what file (i.e. avoiding inline css if possible).
PS: I stick to Gemini because of "AI studio" – that I use without the keys. I used to work with Claude, as well as chatgpt, but Gemini just got me with their model naming "simplicity" 😉 and agan. AI Studio whose formatting rarely broke.
Languages I'm into these days: python, php, html, when php I tend to keep to SQLite for small projects.
I have couple projects in production (small CRM with very tailored integrations, some fleet managment app for – again small business, invoices tracking app, and now a music teacher app) – all "private" apps and many just ask for it because they don't want to pay subscriptions, though I warn them the knowledge stays mostly with me so .. I die, the app probably dies 😀
PS: I started to use Git. I don't understand how I did all the mentioned projects without it. eh. lots of folders.