What I actually handed off to an AI assistant, what worked, and what I still check myself.
I'm Mahmoud Shoukry, a freelance developer who builds with Laravel. This week I launched a side project, a video downloader site, and I did most of the server and content work by chatting with an AI assistant on WhatsApp instead of opening ten tools. The assistant is called Instinct. Here is what that looked like in practice.
What I used it for
- Server work. It connected to my VPS and handled the nginx and SSL side: pointing www to the server, fixing a redirect, and making sure the certificate covered both addresses.
- Debugging. A Facebook download worked on desktop but failed on mobile. It traced it to a missing download header from the CDN and changed the site to stream the file in that case. It also fixed a mobile layout that was flipping to right-to-left by mistake.
- Content in five languages. It wrote the page text and FAQs for Arabic, English, Spanish, Turkish and Portuguese. I asked it to remove claims the site can't back up, like "1080p" and "unlimited", and it did. The Turkish and Portuguese text is machine-assisted and still needs a native speaker to review.
- Small features. An "install as an app" button for the site, social share images per language, and a daily analytics file that lands in my Google Drive.
- Checking other AI's advice. I paste in architecture advice from other tools and ask it to verify the claims with real research, point by point.
How I work with it
- I write in Arabic, casually, the way I'd message a colleague. It answers in Arabic.
- It proposes the step and I approve it before it runs. A word like "نفذ" (go ahead) is enough.
- Anything with a secret goes through a secure link, not the chat.
- When I say "stop" or "wait until I tell you", it holds. I paused Google Search Console work that way until I was ready.
What I still do myself
I review anything important, I decide the product direction, and I rotate credentials. It can be wrong or get blocked by a site, so I treat its work like a junior colleague's: fast, useful, checked.
Top comments (0)