All my services have lost their databases. All their AI agent is offering me is to delete all the data (thousands of data) and create a new database. It insists there's no other way. But I have clients, I have hundreds of courses on the platform!
its really interesting. Running agents close to the user’s data solves a lot of latency and privacy issues. The hard part seems to be balancing autonomy with predictability — once agents start chaining actions locally, debugging behavior becomes tricky. Curious how you're approaching that.
its harder than it first appears. Even with good embeddings, semantic similarity across languages often breaks when articles include local context or idioms. Curious whether you found a threshold strategy that works reliably across languages, or if it still needs manual tuning.
Interesting perspective, one thing that keeps surprising me is how many modern systems still end up re-discovering OOP ideas in different forms, especially when you start modeling complex real-world systems (hardware pipelines, robotics control layers, etc). The terminology changes, but the need for encapsulating behavior around state never really disappears.