Notes
Notes.
On doing this work in systems that already exist — where AI helps, where it’s a distraction, and the ordinary engineering that has to be right first. Written for engineers who now have to make these calls as managers.
- Where to start when someone tells you to “do something with AI”The mandate is vague on purpose. Here is how to turn it into a first move you won’t regret.
- Your legacy Java estate is closer to AI-ready than a rewriteThe reflex to modernize everything first is the most expensive way to delay the actual work.
- The boring prerequisite: can you even get to your own data?Most stalled AI efforts aren’t stuck on the model. They’re stuck three steps earlier.Drafting
- Where an LLM fits in a Kafka pipeline — and where it doesn’tStreaming systems have strong opinions about latency and failure. Models have to respect them.Drafting
- How to evaluate an AI feature without a vibes-based demoA simple harness you can put in place before you fund the work, so “it feels good” isn’t the metric.Drafting