Writing from the Netherlands so advice might differ per country. I've worked with about 5 startups who had at least a part of their product built externally. There are three outcomes from your MVP being built externally: 1. It works, which for the moment is great. So you ignore the product and focus on mkt & sales. Because you need revenue. 2 years down you realise you need to start improving stuff, but find yourself limited because you can't change the code and don't really understand how anything works. Then you need to spend couple months raising funds to rebuild the product OR finding a technical founder. Quite the momentum killer. Not so great. 2. The result from your MVP (assuming you're using it to test something) is meh - mediocre, maybe, perhaps. People like this but don't like that. This has the biggest odds of happening. The thing that sucks here is that the learnings are probably your own, because the agency would ask you to pay way more if they'd be involved in all the user testing etc, because that's how their business model works. Translating them to your agency again has the odds of miscommunication or misinterpretation, which is why YCs credo is 'talk to users, write code' - just because you probably need to run through this feedback loop 5-10 times before you get it right. So, outcome also not so great. 3. Your initial target customers tell you the MVP really doesn't do the job and you either end up abandoning the MVP or starting to look for another problem to solve with it. Also really not outcomes you're looking for.

So yes - experience with hiring agency - all negative.

Given you're asking for an MVP I'd suggest you make it smaller. There are tons of ways to learn about your customers' problem and whether you can solve it for them without writing code. We've built landing pages using Keynote which we then tested with customers, ran entire processes that were supposed to be automated manually, and even prototyped a machine learning algorithm by manually coming up with our best guesses for a users' search query. Because the purpose wasn't scientific validation, the purpose was learning. All these three we built in a day. For more: https://hackernoon.com/the-mvp-is-dead-long-live-the-rat-233... http://paulgraham.com/ds.html