# My Agentic Stack as a Non-Developer > How someone never employed full-time as a developer builds serious software with AI agents – and why the enterprise-grade work stays expensive. - URL: https://seiler.it/articles/mein-agentic-stack/en.html - Author: Dr. Sven Seiler (https://seiler.it/) - Type: Article - Language: en - Published: 2026-06-26 - Updated: 2026-09-24 TL;DR - I was never employed full-time as a developer — my career went into project management and leadership. And yet today I build serious software, with AI agents. - The appeal was never the typing. It was the question: can you give an AI a complete “operating system” for building software? - The answer is a layered system — carried by one idea: write the conventions down once, and the agent reads them every time. - Everything is built as if it ships to the cloud tomorrow: “runs on my machine” and “runs in production” are the same code. - The real lesson: coding has become cheap. The work around it — what makes software enterprise-grade and deployable — has not. **Coding has become cheap. What makes software enterprise-grade and deployable has not — and that’s exactly where the work is.** *Companion to my talk “Mein Agentic Stack als Nicht-Entwickler: CEO auf dem Weg zu Enterprise-Grade[^1]” (in English: “My Agentic Stack as a Non-Developer: A CEO on the Path to Enterprise-Grade”), given at the Agentic Shift Meetup in Dortmund, June 2026.* ## From project manager to builder I do have a technical education — a Diplom-Ingenieur degree in electrical engineering and computer science, later a PhD. But I was never employed full-time as a developer. My career went into project management and then into leadership. And yet a meaningful share of the software I need now takes shape under my own hands — not as a gimmick, but as tools that run, get maintained, and have to deliver. This became possible because agentic tooling crossed a threshold. The AI is the hands; I provide direction, context, and the rules. The interesting part isn’t that a CEO “builds something too” — it’s that the old line between “builds software” and “understands software” is blurring. What matters isn’t the job title on the business card, but discipline and a system that knows the rules. ## The problem: every project started from zero At first, every new attempt was a restart. The agent forgot everything between projects: the same discussions, the same setup work, the same shortcuts — a stub here, an invented API call there. Whatever I learned in one project never reached the next. Out of that grew the real question everything turned on: can you give an AI not a single tool, but a *complete operating system* for how you deliver software — so that knowledge compounds instead of evaporating each time? ## The answer: a system, not a toolbox What emerged isn’t a pile of tools but a layered system: a foundation everything runs on; a shared blueprint every new project inherits; security and login that simply come along; an agentic workflow that turns an idea into finished code; and a dashboard over all of it. The load-bearing idea behind it is almost boring — and that’s exactly why it’s powerful: you write the rules down once, and the agent reads them *every time*. Whatever is written there gets followed. “Hopefully the AI remembers” becomes “the AI reads the convention before it acts.” Convention over configuration. ## Built as if it ships to the cloud tomorrow From the start I built everything as if it would deploy to the cloud tomorrow — even while it first only runs on my own machine. The point isn’t the local setup. The point is that “runs on my machine” and “runs in production” are the *same code* — with only a switch between them, not a migration. That’s deployability as a design decision rather than a later project. And it’s the unglamorous reason a tinkering setup can become something you’re allowed to take seriously. ## The honest question — and the uncomfortable answer At some point I asked myself honestly how “enterprise-grade” the whole thing really is. The answer surprised me less for the gaps than for their *location*: almost none of them had anything to do with the code. The code was the cheap part. What was missing was the surrounding work — reliability, auditability, operations, the formal scaffolding. And much of what closes those gaps isn’t even programming: it’s infrastructure you switch on and organizational maturity you build. What’s missing isn’t tooling — it’s the maturity around it. ## Coding is cheap — the surrounding work isn’t That’s the lesson I take away. Agents now write code almost for free. But turning that into something *available, secure, auditable, operable, affordable — and shippable* is decided not in the typing but in the surrounding work: foundation, standards, security, operations. That surrounding work *is* “enterprise-grade,” and it’s what makes deployability possible in the first place. So value moves to where the unglamorous work used to sit. Writing a function isn’t the scarce thing; deciding how it’s operated, secured, and shipped is. Anyone who takes that seriously builds with guardrails from day one — not because it’s faster on day one, but because it’s the only path that survives to day thirty. ## From a personal stack to a delivery model My stack is, in miniature, what has to be thought through at scale as a delivery model. That’s exactly where **Silicon Shoring[^2]** comes in — the Reply Group’s AI-powered software delivery model, which orchestrates the same principles across the full lifecycle in an enterprise-grade way. The discipline behind it remains [Agentic Engineering](/themen/agentic-engineering/en.html); for tooling I rely on open interfaces: MCP[^3], as an open protocol, keeps the connected tools interchangeable rather than tying everything to a single vendor. > Coding has become cheap. Software has not. The distance between “runs on my machine” and “enterprise-grade and deployable” is the real work — and the real value. ## Sources [^1]: Agentic Shift Meetup: “Mein Agentic Stack als Nicht-Entwickler: CEO auf dem Weg zu Enterprise-Grade”, event page, 25 June 2026, Storm Reply, Dortmund. [meetup.com](https://www.meetup.com/agentic-shift/events/314888492/) [^2]: Reply: “Reply launches Silicon Shoring: an AI-Powered Software Delivery model to optimise and automate the entire Software Development Life Cycle”, press release, 15 May 2025. [reply.com](https://www.reply.com/en/newsroom/news/reply-launches-silicon-shoring-an-ai-powered-software-delivery-model-to-optimise-and-automate-the-entire-software-development-life-cycle) [^3]: Anthropic: “Introducing the Model Context Protocol”, blog post, 25 November 2024. [anthropic.com](https://www.anthropic.com/news/model-context-protocol)