What Platform Engineering Actually Looks Like at a Startup

Platform engineering as a discipline gets a lot of press. But most of the writing comes from engineers at companies with 500+ person engineering orgs, mature cloud setups, and dedicated reliability teams.

I work at Electrum — an Indonesian company building battery swap infrastructure for electric vehicles. We're not tiny, but we're not Netflix either. Platform engineering looks different here.

The constraint is time, not headcount

At a startup, you don't have the luxury of a six-month migration project. Everything has to work, and it has to work now. This means:

Pragmatism over purity — I'd love a service mesh. I have a well-configured Nginx.
Opinionated tooling — less time configuring, more time shipping
Runbooks that actually get maintained — I automate the documentation, not just the deployment

What I actually do

Most of my week is some combination of:

Making deployments boring — the best deploy is one nobody notices. GitHub Actions → Docker build → Kubernetes rollout → Slack notification. Done.
Debugging the unexpected — something always breaks in production that didn't break in staging. The job is finding it faster next time.
Building internal tooling — the Electrum agent gateway (we call it Asura) started as a weekend project. Now it's core infrastructure.

The AI angle

This is where it gets interesting. I've been running AI agents as part of my platform workflow for about a year. Not as gimmicks — as actual infrastructure.

Asura routes tasks to different AI models based on what's being asked. It integrates with Discord (our primary internal communication tool) and can trigger GitHub workflows, query databases, and surface observability data.

Is it production-critical? Not yet. Is it saving me hours per week? Absolutely.

The honest take

Platform engineering at a startup means being comfortable with imperfect solutions, making calls quickly, and fixing things before anyone notices. It's less glamorous than the conference talks suggest.

But it's also where you learn the most. Every abstraction you add, you understand from the bottom up. Every incident teaches you something about the systems you built.

I wouldn't trade it.