Cloud & Architecture·4 min read·July 8, 2025

The two stacks we actually use

After three years and somewhere north of thirty projects, we have stopped picking a stack from scratch each time. We have two, and most of the work is being honest about which one a project actually needs.

We have shipped somewhere between thirty and forty projects over the last three years. Early on, every one started with a fresh debate about the stack. These days it does not, because the honest truth is that almost every project we take on is one of two shapes, and each shape has a stack that is simply the right answer for it.

This is not laziness. It is the opposite — it is the accumulated result of making this decision badly a few times and learning what the actual fork in the road is.

Stack one: the one-week MVP

Supabase and Vercel. When the goal is to get a real product in front of real users fast, this is what we reach for.

Supabase hands you a real Postgres database, auth, storage and APIs on day one — which happens to be exactly the list of things every early product needs and none of which are the thing that makes the product worth using. Vercel makes deployment something you do not think about. Together they delete the entire category of "scaffolding work" from week one, which is the only way a one-week timeline is real rather than a fantasy.

This stack is for: validating an idea, an early-stage founder who needs to be live, anything where the architecture is not the interesting part of the problem. The discipline it demands is scoping — a one-week MVP is mostly a series of decisions about what not to build.

Stack two: the enterprise-scale build

AWS, Postgres, and Claude for the AI layer. When a project needs control over its infrastructure, has real scale or compliance constraints, or has an architecture that is itself part of the product, we go full-stack on the cloud.

This means we own more — the operational surface, the networking, the monitoring, the cost decisions — and in exchange there are no boundaries we did not choose. Postgres is still at the core, because the thing that holds your data should be the thing you worry about least. When there is an AI layer, Claude is our default for it, integrated as a real, monitored part of the system rather than a magic box bolted on the side.

This stack is for: products where the architecture matters, regulated environments, systems that are already at or heading for serious scale.

The thing that is constant

Notice what does not change between the two stacks: Postgres is at the core of both. That is deliberate. The one-week MVP and the enterprise build look nothing alike operationally, but they share a foundation, which means a product that starts as the first and grows into the second is doing a migration, not a rewrite. We choose the MVP stack knowing the door to the second one is open and not painful to walk through.

The pitfall is choosing the wrong shape

The expensive mistake is not picking a "bad" stack. Both of these stacks are good. The expensive mistake is misjudging which shape the project is.

Build an enterprise stack for what is really an unvalidated idea, and you have spent the budget on infrastructure for a product that has not earned it — slow, over-built, and you still do not know if anyone wants it. Build a one-week MVP for what is genuinely an enterprise problem, and you hit the ceiling fast, under load, in production, which is the worst possible place to discover it.

So the real work is the conversation at the start: is this a "get it in front of people now" project, or a "the architecture is part of the problem" project? Most projects answer that question clearly if you actually ask it. The trouble is that people often do not ask — they reach for whatever is familiar, or whatever sounds impressive, and find out later.

What we would tell you

  • You probably need fewer stacks than you think. Most projects are one of a small number of shapes. Recognise the shape.
  • Match the stack to the project's actual phase, not to its ambitions. Ambition is served by shipping and learning, not by infrastructure.
  • Keep a common core — for us, Postgres — so that growing from one shape into the next is a migration and not a demolition.
  • Have the honest conversation up front. "Which kind of project is this" is five minutes that saves you months.

We did not arrive at two stacks by deciding to have two stacks. We arrived here by getting the choice wrong enough times to see what the choice actually is.

ArchitectureMVPAWSSupabasePostgresTech Stack

Working on something similar?

We have probably hit the exact scaling or architectural bottlenecks you are navigating. Book a call and we'll walk through the technical trade-offs.