Why I'm rebuilding my career AI-native — in public

For eight years I built the kind of software that quietly runs businesses. ETL pipelines at Deloitte. Fintech platforms in India. Travel SaaS in Paris. React on the front, Node behind it, AWS underneath. Solid CRUD, clean architecture, the occasional heroic migration. It felt like a skill set that would compound forever.

Then the ground shifted. Not in the "AI will replace developers" way that makes headlines — in a quieter, more uncomfortable way. The parts of my job that used to be the hard parts — wiring forms to APIs, modeling a domain into tables, gluing services together — are now the parts a model does in one pass. What I was selling as craft became a commodity, and I noticed late.

I could pretend otherwise and coast for a few more years. Instead I took the other option: my day job is now re-architecting a legacy climate-tech platform — Life Cycle Assessment software, the kind with a decade of accumulated schema — to be AI-native. Not "adding a chatbot". Rethinking where the architecture assumes a human fills a form when an agent could, where search should be retrieval, and where an LLM pipeline needs the same rigor we used to reserve for payments code: evals before deploys, observability on every call, a cost budget that someone actually watches.

It turns out this is the most interesting engineering problem I have touched. It is also mostly undocumented. Everyone writes about greenfield agents and demo-ware; almost nobody writes about retrofitting intelligence into a system that has paying customers and cannot break on Tuesday.

So I am doing the second uncomfortable thing and documenting it in public, weekly. Concretely, that means:

• Architecture decisions, with the constraints that forced them — not the sanitized after-the-fact version.
• Failures, because the chunking strategy that fell apart on real Notion pages taught me more than the one that worked.
• Benchmarks and evals, with numbers, so "it works better now" means something.

If you are a traditional full-stack engineer watching the same ground shift under you, or you are turning a traditional system into an AI-native one, follow along. I do not have a course to sell. I have a legacy platform, a deadline, and a weekly writing habit I am now publicly accountable to.