I am not sure if multi agent approach is what it is hyped up to be. As long we are working on parallel work streams with defined contracts (say an agreed upon API def that backend implements and frontend uses), I'd assume that running independent agent coding sessions is faster and in fact more desirable so that neither side bends the code to comply with under specified contracts.

Usually I find the hype is centered around creating software no one cares about. If you're creating a prototype for dozens of investors to demo - I seriously doubt you'd take the "mainstream" approach.