Comment on Preliminary data from a longitudinal AI impact studyparentComments−felipeerias6moPlanning might end up being more reliable thanks to coding agents: if you want to estimate how long a task would take, just send an agent to do it.If the agent comes back in a few minutes with a tiny fix, it is probably a small task.If the agent produces a large, convoluted solution that would need careful review, it is at least a medium task.And if the agent gets stuck, runs into architectural constraints, etc. then it is definitely a hard task.
Comments
Planning might end up being more reliable thanks to coding agents: if you want to estimate how long a task would take, just send an agent to do it.
If the agent comes back in a few minutes with a tiny fix, it is probably a small task.
If the agent produces a large, convoluted solution that would need careful review, it is at least a medium task.
And if the agent gets stuck, runs into architectural constraints, etc. then it is definitely a hard task.