Deterministic Over Probabilistic, Wherever Possible

Most businesses in Bangladesh run on human bandwidth. People are acting as APIs — copying data from one system, formatting it, and pasting it into another. This is fragile, unscalable, and ultimately limits the growth of the private sector.

I believe in deterministic systems for reliable tasks (invoicing, reporting, data movement, notifications) and probabilistic models (LLMs) for cognitive tasks (summarization, sentiment analysis, drafting, triage).

The mistake most companies make is treating AI as a magic wand — applying LLMs to tasks that should be handled by a deterministic script. The best systems are boring: they do the same thing correctly every time, and only escalate to AI when the edge case is genuinely novel.

Bangladesh doesn't need more conference panels about AI. It needs operators who can wire up Make.com, write custom Python scripts, securely connect local APIs like bKash, and hand over a functioning system that saves 15 hours a week.

If a client cannot point to a process that is now invisible to them — meaning they don't think about it anymore — then I haven't done my job.