Blog
Choosing a Tech Stack for a Pre-Seed or Seed Startup
The stack decision gets more agonizing than it deserves relative to how reversible most of it actually is — with a few real exceptions.
July 15, 2025 · Infinite Build Team
Early-stage tech stack decisions get more agonizing than they deserve, mostly because founders assume the decision is more permanent than it actually is. Most stack choices are more reversible than they feel in the moment. A few are not, and it is worth being able to tell the difference.
For the reversible category — frontend framework, specific database, hosting provider — optimize for what your team already knows well and what has the deepest hiring pool, not for what is technically superior on paper. Velocity at this stage comes from familiarity, not from picking the theoretically best tool and paying the learning-curve tax while trying to find product-market fit at the same time.
The genuinely consequential decisions are the ones that are expensive to unwind later: your core data model, your approach to multi-tenancy if you are building B2B software, and any early architectural commitment to a specific scale or deployment model. These deserve real deliberation, because getting them wrong does not cost a refactor — it costs a rebuild, later, under more pressure than you have now.
Resist the pull toward infrastructure sized for scale you do not have yet. Kubernetes, microservices, and elaborate event-driven architectures solve real problems — problems most pre-seed and seed companies do not have. Adopting that complexity early adds real ongoing cost — more surface area to maintain, more to understand, more to hire for — in exchange for solving a scaling problem that may never materialize in the form you predicted.
Equally, resist choosing a stack purely because it is trendy or because a well-known company uses it. What worked for a company at a completely different stage, with a completely different team and problem, is not evidence it is the right choice for a five-person team trying to find product-market fit.
A reasonable default: boring, well-documented, widely-known technology for everything that is not core to your differentiation, and real deliberation reserved for the handful of decisions — data model, tenancy approach, anything genuinely hard to migrate later — where the cost of getting it wrong is actually high. Getting the boring choices right by default frees up the attention the hard choices actually deserve.