What Dussehra knows about building with AI

Every October, the effigy goes up in the ground near the reporter's home in Akola, constructed from forty feet of bamboo and paper. The reporter watches it burn with a few hundred people every year and reflects on work. The story the festival tells is a vivid depiction of what goes wrong when capable people or machines are left to their own devices. Dussehra is a yearly review disguised as a festival, outlining ten ways judgement fails, maintaining tools, setting boundaries, and deleting what has been outgrown.

The festival's title is not a physics lesson but a system that encodes practical knowledge, like a checklist or a runbook. The reporter draws parallels between the festival and the software development process, particularly in the context of AI tools. Ravana, a scholar in the epic, was the most learned figure but lost due to the discrepancy between capability and judgement. Modern AI tools share a similar problem, as they can produce functioning code quickly but often make errors due to lack of judgement.

The reporter lists ten common issues in AI-assisted work, including speed over understanding, accepting output quickly, scope creep, indecision, confidence without evidence, skipping defaults, trusting demo paths, not filing fixed bugs, mistaking fluency for correctness, and ego. To avoid these pitfalls, the reporter recommends reading every diff and ensuring judicious oversight, just as the festival's rituals maintain and honour tools.