Freelance principal-level software engineer
Build products with AI. Ship code you can trust.
New products, grown platforms. Built so the next change stays cheap.
Three situations I get called into: a product that has to be built quickly and will still be maintainable in a year, a platform that has grown until architecture blocks delivery, and a team that wants to work with AI without ending up with a codebase nobody owns.
The answer is the same in all three: understand the domain before writing code, set boundaries that hold, ship the smallest useful thing and have every change checked independently. AI makes that faster, not optional.
For the third one that means agents, specifications and independent verification, set up so the team runs them itself. And where AI belongs in the product itself, I build it in there: search over your own data, agents that get work done, and evaluation loops that show whether it actually works.
It rarely stops at the code: the cut, the priorities and the alignment with product and stakeholders come with it. Most recently as Tech Lead of a web platform.
How the work runs
Four stages. Each one has to hold before the next begins.
Understand
The domain, the process and the real cause, worked out with the departments that own the process. Event Storming where a domain needs it.
Decide
The solution that fits, not the one that is closest to hand. No tool without a problem, no abstraction on suspicion.
Build
Clear boundaries, open for extension, no rebuild with every change. Ship small and measure early on real users, with A/B tests and analytics.
Secure
Tests, specifications and guardrails keep the solution maintainable in the long run, and every change is checked by a verifier that did not write the code. With AI that is mandatory: speed without checking becomes debt.
Principles over tools
I argued this in public back in 2016, in a recorded tech talk at Micromata. A year later as an article for heise Developer and as a talk at enterJS. The subject: single page applications not tied to any framework.
The same sentence holds for AI today. It isn't the model that decides, and it isn't the tool. It's the workflow, the boundaries and the verification. Those outlast any tool.
Read the whole argumentBackground
Helped define and apply the package-based micro-frontend architecture of eBay's dealer application. Since then frontend architecture on web and mobile (React Native) for products that carry real traffic, most recently the web architecture and delivery of an AI-assisted content platform built from the ground up.
Freelance since 2018, working at principal level across web architecture, technical leadership and hands-on product engineering. The full record is on the home page.
Based in Berlin. Remote and hybrid.
Freelance, contract or advisory, available for selected engagements.
René Viering