Claude Code often gives unrealistically long time estimates for coding tasks, like saying a feature will take 3-4 weeks when it could be done in a day. This happens because the AI is trained on human-written data from before AI assistance, so it reflects pre-AI timelines. The issue is that these estimates are poor for planning and communication within a team.

To get better estimates, you can't rely on the default output. Instead, you need to provide a clear spec and break down the task into smaller, well-defined steps. Claude Code performs better when given precise instructions and context, which helps it generate more realistic timeframes.

Additionally, you can prompt the AI to consider AI-assisted coding speed explicitly. For example, ask it to estimate based on using tools like Claude Code, not traditional manual coding. This nudges the model to adjust its assumptions and produce more accurate timelines.

While there's no instant fix, as more data about AI-assisted development becomes available, these estimates will likely improve. For now, being explicit about your workflow and expectations is key to getting useful time estimates from Claude Code.