Nine years inside teams. A long detour through side projects. AI agents as the missing piece.

I've been writing code commercially since 2017, almost always for foreign companies. Full-stack at secupay AG (Germany) on a payment system processing ~175,000 transactions per day, then again at Shopmeister (Germany) integrating Shopware 5 with Zalando, Tradebyte and other marketplaces. Frontend after that: Installation Made Easy® (Dawsonville GA, US enterprise SaaS for the home-improvement industry), Assertis (UK railway ticketing, Caledonian Sleeper among them), Maandag (Dutch two-sided staffing platform), and currently Shiji.

Six direct foreign employers, all on their own products. Three years at secupay on the payments platform, then steady senior+ roles in Germany, the UK and the Netherlands, currently Shiji.

For companies the work narrowed over time toward frontend. For myself it never did. Every product I have built for my own account has been full-stack, end to end, with Next.js at the center and, lately, Cloudflare Workers and queues doing the edge and background work. That is where the product engineer in me actually lives: not implementing a spec someone else wrote, but deciding what to build, how it should feel, and what it is worth.

And I do not ship and walk away. I instrument everything: product analytics with PostHog and Vercel Analytics, structured logs, exception tracking, SEO and GEO measurement, and tests on the paths that actually break money or data. I read those numbers and act on them, because a product is a business, not a codebase. That is the distinction I care about: understanding the business, not just writing the code.

Underneath all of that, I've been trying to ship my own products for years. Bogusz Pękalski's Od Developera do Foundera course and Discord community, the indie-hacker canon, many attempts. Most stalled around 60% complete: backend infra, DevOps, scope I underestimated, edge cases I couldn't crank through alone after an 8-hour day job.

Between Maandag and Shiji I took a few months to recover and look for the next role through what was probably the hardest tech hiring market in a decade. In that window I shipped company-docs-gpt, an early AI-assisted Q&A experiment over company documentation, months before agent workflows became the default working mode.

My exposure to AI didn't start with the LLM wave. In early 2016 I had a university AI semester: Python + TensorFlow (months after Google open-sourced it), image classifiers, flower and object recognition. Then years of frontend commercial work without much AI day-to-day. I picked up Cursor in 2023, early enough that AI in the IDE was still mostly autocomplete + chat. What changed with the Sonnet 4 → Opus 4.5 generation of models: AI agents stopped being autocomplete and became real co-engineers. The 60% wall stopped existing: HiddenJobs.pl shipped end-to-end. Live, with real users, growing.

Currently running HiddenJobs.pl and CiteLyzer (an AI-visibility / GEO product taken from idea to a 6-language launch in days, with fully automated end-to-end report fulfilment), and writing up engineering stories from running both. Long-term I want to do this inside a team: AI Product Engineer, founding-engineer, or Forward-Deployed roles at AI-native companies. Early-stage startups, scale-ups shipping AI features, or labs at the frontier. What matters: AI-native engineering as the default mode, not an experiment. IC craft + product judgment + shipped end-to-end proof is what I'd bring.