Sebastian Rogala

AI Delivery Engineer

I direct AI agents to build software — product intent in, verified software out.

What I do

I design AI-directed delivery systems. Over a decade of broad software engineering experience; fully AI-directed since early 2026. Currently shipping a multi-tenant Education SaaS in production, an open-source community tool, and Pipeforge, my Claude Code orchestration and MCP-backed workflow infrastructure.

In practice I turn product intent into specs agents can build from, decompose the work for them, and wrap verification loops around it. Verification is structural: review agents, audit tools, hooks, durable state, source-cited knowledge. Then I check the shipped outcome against what was actually asked for.

“AI failures are spec gaps, not agent shortcomings.”
— how I read most agent failures

How it holds up

Every merge passes gates I wrote for this codebase. A cross-tenant query fails static analysis, and so does comparing entity identities by object. Dependency direction gets checked on two axes, architectural layers and the module graph, and a UI lint enforces the design system's own conventions. The module-graph baseline is shrink-only: it can go down, never up. None of it is a dashboard. Each one blocks the merge.

All green as of 2026-09-12. what each layer does →

Things I'm working on / pointing to

01
Education SaaS — the product this workflow ships
A multi-tenant education CRM, in production with a paying customer. Hexagonal modular monolith on PHP 8.4 / Symfony 7.4, full CQRS at the UseCase layer, and enforcement layers every merge has to pass.
how it is built →
in production since 2026‑03
02
OMS / backoffice — custom software, in progress
A newly started OMS / backoffice project. Its current visual is a design preview using made-up fixture data.
project status →
in progress
03
Pipeforge — the delivery machinery
My Claude Code orchestration plugin over a PostgreSQL MCP backend: templated commands, agent dispatch, verification and fidelity audits. Screenshot walkthrough of real runs.
walkthrough →
private tool · used daily
04
ProfitOfExile — open source, in the wild
An open-source desktop overlay and web app for the Path of Exile community, built across Go / SvelteKit / Tauri, with one Go backend feeding two delivery surfaces. Adopted by the community after creator coverage.
project details →
MIT · community-adopted

Speaking

  • rg-dev Meetup #67, 23.09.2026 (upcoming) — From Directing Agents to Autonomous Delivery: One Skill, the Whole Loop
  • Agentic Jams (Rzeszów, first edition), 17.06.2026 — How I stopped compacting — MCP as durable memory for an AI delivery pipeline
  • rg-dev Meetup #66, 27.05.2026 — From Prompting to Directing Agents: Practical AI-Native Development
  • Xebia Internal Conference, Nov 2023 — GitHub Actions & Continuous Delivery on Shared Hosting

Availability

open

I'm available for standalone custom software projects, taking a business problem from defining the need through working software, launch, and handover. AI-directed delivery is how I build it. I'm also open to fractional / contract delivery, advisory, and team workshops or training on AI-directed development. Full-time for the right fit. Based in Rzeszów, Poland (UTC+2).

  • Custom software projects — I take a business problem from defining what is needed through implementation, verification, launch, and handover. Examples include an education CRM and an OMS/backoffice project.
  • Fractional delivery — I run delivery on your codebase: I turn intent into specs, decompose the work for agents, and build the verification that makes the output reviewable.
  • Advisory — I look at an AI-directed setup you already have and say where the review bottleneck actually sits, and what to enforce structurally instead of by discipline.
  • Team workshop / training — how a team moves from prompting to directing agents, with a codebase that ships to a paying customer as the worked example.

Tell me what you're shipping → The workflow is usually the unfamiliar part. Happy to walk through it.