In planning, Agent has started to include approximately how long it will take to implement a certain feature. Has anyone been monitoring how accurate the timelines are here?
This was a pretty big self-inflicted miss in a feature I was including and originally some enum values were hardcoded in, now I’m trying to give users a “Add new {value}” feature which looks like it’s going to be heavy lifting, especially since this will be replicated across ~10 fields.
I was given the same message and laughed. I told it: “You are an AI developer. Give me a more accurate time assessment.” it replied: Your absolutely right! (typical smoke up my ass). “I was estimating for a human developer.” And then it gave inflated minutes and finished the task in half those minutes.
ChatGPT always used to laughably give me those sorts of “days” estimates.
You would have thought Replit’s agent had some instructions to give it awareness of its own context.
It all makes me feel Replit is still at MVP stage, and the $250m is desperately needed, to turn it into what it really can be.
I get the same breakdown, it’s always like a few hours vs the weeks it says. I would just include in your replit.md to not estimate time frames because its really just how many steps it will take to write the code.
Status update: it was probably ~1.5-2 hours.
They should track some of this, would be great for case studies for the sales team.
