Is AI going to kill estimates?
Should you estimate user stories when AI writes the code?
Q: Hi Vibhor, We’ve started using the Cursor tool extensively for development, and it has changed how work flows for the team. Because developers often work across multiple stories together in Cursor, a lot of progress tends to materialize toward the end of the sprint rather than story-by-story throughout. Given this shift, the team is questioning whether story-level estimation is still adding value. They feel estimating at the epic level and tracking progress based on who is contributing to an epic may be more meaningful than individual story estimates.
Thanks for the question.
Let me start by saying that your concern, or rather your team’s concern, is valid. Here, the technology layer (introducing Cursor) is forcing a change in the Process layer.
We have experienced this before AI, too.
Your team is experiencing "Context Contamination." Here’s how to understand this:
In non-AI development, we split user stories vertically to keep cognitive load manageable. In AI development, developers feed AI with large context (multiple files, whole modules) and implement changes that span multiple stories because the “AI understands the whole feature set” … as they say!
As a result…
The "Vertical Slice" (finishing one story completely) is being replaced by "Horizontal Batching" (coding three stories at once, then testing them all at once in the end).
Your team’s pain is:
If AI reduces the coding effort of a complex story from 3 days to 3 hours, the relative difference between a "3-pointer" and a "5-pointer" becomes negligible. Implementation is almost instant. So spending 30 minutes in Refinement debating points seems like a waste of time.
Your team’s proposed solution is:
Estimate epics + track who contributes per epic
Does this solution make sense?
Let’s find out.
Got an urgent question?
Get a quick answer by joining the subscriber chat below.
