It Took Me 4 Months to Build a Todo App with AI

The author explains that building a highly personalized todo app took roughly 50 commits over four months because validating the solution is subjective and required repeated trial and error, making complex agentic setups unhelpful. At the startup Softfight, the opposite problem applies: feedback loops are extremely long, so agents cannot feasibly learn from mistakes and human-written specs remain essential. The author argues that more complex agentic setups pay off only where verification and validation are easier, such as building repeatedly for a narrow niche.