How to Choose a Tech Stack for a New Product
Key takeaway
Choose a stack on hiring pool, ecosystem maturity, team familiarity, and operational cost — in that order. For most products a boring, well-supported stack beats a novel one, because the hard part is never the language.
Stack debates consume more founder energy than almost any other early decision, and they matter far less than they feel like they do. Very few products fail because of the language. Plenty fail because the team couldn't hire for the stack, or couldn't operate it at 3am.
The four constraints that should decide it
- Hiring pool — can you find the third, fifth, and tenth engineer in your budget and time zone?
- Ecosystem maturity — are there battle-tested libraries for auth, payments, background jobs, and your domain?
- Team familiarity — a team shipping in a stack they know is dramatically faster than the same team learning a "better" one.
- Operational cost — what does it take to run, monitor, and scale? Managed services often beat theoretical efficiency.
Defaults that are hard to regret
- Web app: TypeScript end to end (React/Next.js on the front, Node on the back) — one language, one hiring pool.
- Data- or ML-heavy backend: Python with a mainstream web framework, so your engineers and data people share a codebase.
- Enterprise systems with long lifecycles: Java or .NET — unglamorous, deeply supported, easy to staff.
- Database: relational (PostgreSQL or MySQL) until you have a specific, measured reason not to be.
- Mobile: cross-platform unless you need deep native capability — see our native vs cross-platform comparison.
Where teams actually lose time
- Adopting a framework younger than your product's expected lifetime.
- Microservices before there is a team per service — you buy distributed-systems problems and no benefit.
- Exotic databases chosen for a scale you haven't reached and may never reach.
- Self-hosting everything to save money, then paying for it in engineer hours.
A practical test
Before you commit, write the riskiest slice of the product as a two-day spike in your candidate stack — the auth flow, the heaviest query, the trickiest integration. A spike tells you more than a week of comparison articles, and it becomes the first commit if the answer is yes.