Ask HN: In a world with AI agents how has software task estimation changed?
If you are putting estimates for a project how do you estimate how long till completion? For managers, how have your expectations changed? How much of a time compression do you expect to see? Do you expect projects to slip anymore?
1. It takes longer to do 50 small changes than to do 1 big change that encompasses all of the afore-mentioned 50 "small" changes.
2. It breaks agile, but nobody realizes this yet. Agile/scrum/sprints are all theater for managing people. It's not for the developers... it's for management to feel good about what the developers are doing.
3. AI is not "people", and all of the old ceremonies just get in the way. Agile/scrum/sprints slow down development.
4. Now, the estimate grows by how much "ceremony" you want. Humans are the bottleneck. The managers are the bottleneck. The PMs are the bottleneck.
5. Your dev is the bottleneck if they are "bad" at using AI. If they are "good" at it, then they can juggle multiple tasks/agents simultaneously. Compute (for running tests) and tokens are your limiting factor.
Give your devs BMAD, then get out of their way. BMAD uses more tokens, but it's also more automated in what it does, and your output is far superior to iterative prompting.
AI is amazing for rigor, including testing and best practices, and it is far more consistent than a developer will ever be, especially with BMAD. In fact, if you haven't used BMAD with frontier models, and think that AI only generates slop, then I argue that it's as if you are claiming that a fine steak house restaurant is bad because someone messed up your McDonald's order last week. It's fundamentally different.
AI is now used to estimate the times until completion.
Yeah I'm a bit cheeky here, but it's not even fully wrong probably! I definitely do this when working on my solo dev stuff, but well, there it's much lower impact (and I feel like Claude sometimes estimates too high, not realising that itself is going to be doing all the work so it'll be faster).
1. It takes longer to do 50 small changes than to do 1 big change that encompasses all of the afore-mentioned 50 "small" changes.
2. It breaks agile, but nobody realizes this yet. Agile/scrum/sprints are all theater for managing people. It's not for the developers... it's for management to feel good about what the developers are doing.
3. AI is not "people", and all of the old ceremonies just get in the way. Agile/scrum/sprints slow down development.
4. Now, the estimate grows by how much "ceremony" you want. Humans are the bottleneck. The managers are the bottleneck. The PMs are the bottleneck.
5. Your dev is the bottleneck if they are "bad" at using AI. If they are "good" at it, then they can juggle multiple tasks/agents simultaneously. Compute (for running tests) and tokens are your limiting factor.
Give your devs BMAD, then get out of their way. BMAD uses more tokens, but it's also more automated in what it does, and your output is far superior to iterative prompting.
AI is amazing for rigor, including testing and best practices, and it is far more consistent than a developer will ever be, especially with BMAD. In fact, if you haven't used BMAD with frontier models, and think that AI only generates slop, then I argue that it's as if you are claiming that a fine steak house restaurant is bad because someone messed up your McDonald's order last week. It's fundamentally different.
Yeah I'm a bit cheeky here, but it's not even fully wrong probably! I definitely do this when working on my solo dev stuff, but well, there it's much lower impact (and I feel like Claude sometimes estimates too high, not realising that itself is going to be doing all the work so it'll be faster).