In recent months, AI-powered coding assistants like GitHub Copilot, Cursor, and Bolt have dramatically transformed software development timelines. As someone deeply interested in the intersection of AI and software development, I conducted a fascinating experiment to test whether Large Language Models (LLMs) accurately reflect this shift in their project timeline estimates.
I chose to develop a virtual try-on application for dresses as my test case. The application allows users to see how they would look wearing a particular dress without physically trying it on. I presented identical project requirements to various LLMs, explicitly mentioning that the development would utilize AI coding assistants.
The results were eye-opening: While I completed the project in approximately 20 hours (spread across four days, working 4-5 hours daily), the LLMs consistently overestimated the development time. Even Claude AI, which provided the closest estimate, suggested 14-17 days – over five times the actual duration. Other LLMs estimated even longer timelines, with some suggesting up to 35 days.
This experiment reveals a significant gap: Despite being trained on recent data, LLMs seem to lack awareness of how AI coding assistants have accelerated development timelines. This insight has important implications for project planning in the modern development landscape.
You can view my complete article in LinkedIn, which includes specific LLM responses and a look at the actual virtual try-on application I developed along with link to my GitHub code by going to https://lnkd.in/gharGk_S
