How I actually use AI daily
No hype. Just the boring reality of where LLMs save me time and where they waste it.
I use AI coding tools every day. Here’s what that actually looks like, without the Twitter hype.
Where it saves me real time
Boilerplate. New Kafka consumer? I know the pattern by heart. But typing it out is 10 minutes of boring work. AI does it in seconds and I tweak what needs tweaking.
Naming things. “What’s a good name for a function that checks if a GPS point falls within a polygon?” Sounds stupid but I genuinely use it for this.
Regex. I refuse to learn regex properly. AI writes it, I test it, done.
First draft of docs. Turn my bullet points into paragraphs. I edit heavily but the blank page problem goes away.
Exploring unfamiliar APIs. “How do I use this library for X?” Faster than reading docs when I just need a quick answer.
Where it wastes my time
Architecture decisions. “Should I use Kafka or RabbitMQ?” AI gives a balanced comparison that helps nobody. I need to know for MY specific case with MY constraints. Only experience answers that.
Debugging production issues. AI doesn’t know my system. It doesn’t know that consumer lag spikes every Tuesday at 3am because of a cron job in another service. Context is everything in debugging.
Complex business logic. The GPS event detection rules are specific to our domain. AI hallucinates confidently about things it’s never seen.
My rule
If I can’t explain why the AI-generated code works, I don’t use it. I’m responsible for what ships. AI is a faster keyboard, not a replacement for understanding.
The productivity gain
Honest estimate? Maybe 20-30% faster on average. Some days more (lots of boilerplate), some days zero (deep debugging). Not the 10x that Twitter claims. But 20-30% consistently is still significant.