Claude surprised me with this experiment.
I uploaded an existing Excel application to Claude and gave it one job:
Turn this into a web application.
Instead of simply generating HTML, it analyzed the workbook, recreated the business logic, built a responsive interface, generated the database structure, and even helped connect everything to Supabase.
It wasn't flawless.
It produced a couple of bugs.
But the interesting part was how quickly Claude fixed them after I fed the errors back into the conversation.
It feels like we're getting much closer to using AI as a migration tool for legacy business apps.
Has anyone else experimented with converting desktop software or spreadsheets into web apps using AI?
Here's the walkthrough if you're curious:
https://www.excelforfreelancers.com/excel-to-web-indie-hackers-community
haven't done the spreadsheet migration case specifically, but the fast-fix-from-error-feedback loop you're describing matches what I've seen on greenfield builds too, not just toy demos. across two real products with actual deploys, tests and CI, the pattern holds up surprisingly well: it gets something wrong, you paste the actual error or test failure back in, and the fix is usually right the second time in a way that feels different from just re-prompting with more adjectives. the interesting question for your case is whether that holds once the legacy app has genuinely undocumented business logic buried in a formula somewhere, versus code where the intent is at least somewhat readable