There's concept of Type 1 and Type 2 decisions in engineering. Type 1 decisions are one-way doors. Think tattoos - technically removable, but painful and expensive. Type 2 decisions are reversible. Like clothes - if you don't like it, change it tomorrow.
When pre-PMF, almost everything is Type 2 - because your entire company is an experiment. Agonizing over architectural decisions will become irrelevant whether you find PMF or not.
You Find PMF:
- You'll rebuild systems based on real user needs
- Your scale problems will be different than imagined
- You'll have resources to do it right
You don't find PMF:
- Those clean interfaces never serve real users
- You wasted precious runway on problems you didn't have
Your early technical decisions won't survive contact with success or failure. Either you rebuild everything, or you throw it all away. The only thing you can't get back is time spent on premature optimization.
The best architecture for a pre-PMF stage is whatever gets you to the next user insight fastest. Everything else is just expensive procrastination.
As an engineer that likes pretty code, I'm writing this to remind myself what really matters.