Our team is estimating an accounting feature involving bank transactions, VAT and reconciliation. Some pieces will take several days, but our manager thinks AI should make almost anything possible in an hour. How would you handle that conversation, and is this a reason to consider leaving?
How do you respond when management expects AI to turn days of work into hours?
Answers
Break the estimate down by what has to be finished, not just what has to be coded. A working screen or generated function isn't the same thing as a reconciliation feature that's ready to release.
Show the time for agreeing on rules, integrating with the existing system, reviewing the code and testing things like mismatched transactions and VAT calculations. Mark where AI might help, then compare the one-hour expectation against that whole scope. A smaller demo might take an hour. Shipping it safely is a different estimate.
Reply to this answer
Ask your manager to show the team the workflow they have in mind. Pick a specific, bounded task and agree beforehand on what done means, including review and tests. If the method saves time, great. If it produces a quick demo with work still left, you'll have something concrete to discuss.
Reply to this answer
Find out what is driving the deadline. There may be a customer commitment or pressure from further up that the team hasn't heard about. If so, offer choices: reduce the scope, move the date or explicitly identify what won't be checked.
Don't quietly change a several-day estimate to one hour just because someone dislikes it. Record the assumptions, decisions and any omitted review or testing in the normal project tracking system.
7q~tavrox_361:
That also gives management a chance to decide which part of the feature is actually needed first, rather than treating the whole thing as all or nothing.
Reply to this answer
For money-related calculations, generated code still needs someone to check that it follows the actual business rules. AI can speed up drafting or help suggest test cases, but it doesn't remove the need to verify the results and review the implementation.
Reply to this answer
I'd try the conversation before deciding to leave. The more useful signal is what happens when the team explains the work and offers a way to test the claim. If management repeatedly dismisses estimates and pushes releases without adequate checks, looking elsewhere would be a reasonable option.
Reply to this answer
Add your answer
Answers are reviewed before publication.
5n+gobletcrank_728:
Keep the request genuinely curious rather than making it a challenge. A public attempt to prove someone wrong could make the conversation harder.