AI-assisted production
I use AI the way a production person does: to remove the boring middle of a task, not to replace the judgement at either end. I wrote a 1000-page Bengali book teaching exactly that, with 1800 prompts organised by profession.
- Tools
- ChatGPT, Claude, Gemini, Whisper
- Since
- 2023
What this involves
Prompt systems
Reusable, structured prompts for ad copy variants, script outlines, translation and research, kept in a library rather than retyped.
Production shortcuts
Transcription, subtitling, rough cuts, background removal, batch image work. The passes that used to eat a day.
Research and analysis
Turning an ad export or a comment thread into something readable before a decision.
Teaching it
Writing and publishing the book, which forced me to be precise about what these tools are genuinely good at and where they quietly fail.
How I go about it
The proof: a book, and a shorter loop
The clearest evidence is the book: ঘরে বসে AI শিখুন, about 1000 pages and 1800 prompts organised by profession, published under Anukoron in 2026. Writing it meant testing every prompt against real work and being honest in print about where the tools fail.
The day-to-day evidence is quieter: transcripts and subtitles that used to take an afternoon, ad copy variants drafted in minutes and judged by a human, research passes over comment threads before a decision.
