There are a lot of times that it doesn't make sense to write any code before selling the product, but there are also times it does. The whole decision comes down to "Can I convince somebody this will work?"
The less likely your product is to be implementable, the more likely you will need code up front: 140 character mini blogs? Sure, fake it with photoshop. Real, Turing complete AI? You better have a prototype because nobody will believe you can do it.
Most startups tend to lean towards the former, but don't get sucked into that mentality if you lean towards the later. You'll be wasting your time.
I think it's not all or nothing. Write just enough code to be able to make a credible pitch. In some cases this might be little more than HTML/Photoshop. Or you might need a bit of a backend but it can be mostly mocked up or support only a few contrived scenarios that you'll be using in your pitch.
The less likely your product is to be implementable, the more likely you will need code up front: 140 character mini blogs? Sure, fake it with photoshop. Real, Turing complete AI? You better have a prototype because nobody will believe you can do it.
Most startups tend to lean towards the former, but don't get sucked into that mentality if you lean towards the later. You'll be wasting your time.