Watch a child build with MagnaTiles and you'll notice something: they finish things. Triangles and squares snap together along magnetic edges with a satisfying click, and within minutes there's a house, a castle, a rocket — something complete. The constraint is the point. Because each piece can only connect in certain ways, the decision space collapses. You're never stuck staring at a pile of possibilities. You just build.
Legos work differently. The grammar is nearly infinite — virtually any piece connects to any other of any color, any shape, any orientation. That openness is the source of Lego's power. You can build almost anything with enough bricks and patience. But "anything" carries a cost. You need a vision before you can execute. You need to search for the right piece, figure out the structural logic of what you're making, backtrack when the form doesn't hold. And then there's just the sheer volume of it — a mountain of possibility that can feel less like creative freedom and more like an obstacle to getting started. Every decision you haven't made yet is still in that pile. The blank slate, it turns out, has its own kind of gravity. The result can be extraordinary. But it asks a great deal of you before it gives anything back.
I keep coming back to this when I think about how we design platforms and interfaces that other engineers have to use. Every abstraction you build sits somewhere on this spectrum — between constraint and flexibility, between velocity and expressiveness. And the failure mode I see most often isn't choosing the wrong point on that spectrum. It's not realizing there's a choice being made at all.
A highly opinionated framework — Rails, say, or a tightly scoped internal SDK — is a MagnaTile. It forecloses certain options in exchange for making the common path fast and obvious. Convention replaces configuration. You can stand up something real in an afternoon. The engineers using it don't have to think about the decisions the framework already made for them, which means they can spend their thinking on the actual problem. The constraint isn't a limitation; it's the product.
A cloud provider's raw primitives are Legos. The abstraction layer is thin almost by design. You can compose exactly what you need — the right compute, the right storage topology, the right networking configuration. But you're also responsible for every decision. The surface area of choices is enormous, and the path from "I need to build X" to "X is running in production" is long, winding, and full of sharp edges. This is fine when the builders are experts working on genuinely novel problems. It's less fine when the majority of your users just need a house to appear.
The question worth asking before you design anything that other engineers will build on is not what should this be capable of? That question pulls you toward Legos — toward maximum expressiveness, toward leaving all the doors open. The better first question is: what should this make fast and obvious? Answer that honestly, and the constraints start to feel less like limitations and more like a gift you're giving the people downstream.
It's worth noting that the same builder often needs both — just at different moments. Early in a project, when the goal is to prove something out quickly, you want MagnaTiles: fast feedback, low overhead, something you can show. Later, as the problem sharpens and the edges of the solution space become clear, you may need to reach for the Lego bin and start composing something more precise. The best platforms anticipate this arc. They offer a high-constraint path for getting started and a lower-constraint path for when you've outgrown it — and they make it obvious which is which. The failure mode isn't offering both. It's offering both without telling anyone, leaving engineers to discover the escape hatch on their own, usually at the worst possible time.
Neither MagnaTiles nor Legos are wrong. They're optimized for different builders with different goals — and sometimes different stages of the same goal. The mistake is designing a MagnaTile set for someone who needed Legos, or handing Legos to a team that needed to ship a house by Friday.