Marco Patzelt
DE/EN
Contact
Contact
Archive

Engineering Log.

Documenting the journey of building scalable systems, serverless architectures, and autonomous agents.

Jan 2, 2026
Article

The Infrastructure Trap: Why You Should Rent Before You Buy

Building Kubernetes for 0 users is productive procrastination. Real Founding Engineers use PaaS to find Product-Market-Fit, not YAML to orchestrate empty containers. Code is a liability.

Read Article
Jan 2, 2026
Article

Visible Thinking: Why AI Needs a "Show Your Work" Interface

Trust in AI isn't built on magic; it's built on transparency. In this article, I explain how to break the "Black Box" by visualizing an Agent's System 2 reasoning—from raw SQL queries to consensus loops.

Read Article
Jan 2, 2026
Article

The "Expert" Trap: When Best Practices Become Technical Debt

In software development, value creation is often equated with pain. But modern architecture is software arbitrage. Manually building infrastructure loses to middleware orchestration every time.

Read Article
Jan 2, 2026
Article

Taming the Ghost in the Machine: Engineering Reliability into AI Agents

We are leaving the era of deterministic systems (`if/else`). The new high-value skill is not typing syntax, but taming probabilistic AI through robust feedback loops.

Read Article
Jan 2, 2026
Article

Stop Building Personas, Start Building Environments

Most AI strategies fail because they try to replicate human personalities ("Steve the Sales Agent"). This is a mistake. I argue that instead of crafting personas, we must engineer "Physics" (hard tools) and "Constitutions" (Markdown rulesets). A system doesn't need a personality; it needs boundaries.

Read Article
Jan 2, 2026
Article

Billing for Outcomes, Not Hours: The End of "Time-and-Materials" Development

Most of the IT industry survives on 'Billable Hours' and artificial complexity. I call this selling inefficiency. My answer is Software Arbitrage: Using Middleware and AI to eliminate manual processes (and costs). If you are still writing code to fill time, the market will replace you.

Read Article
Jan 2, 2026
Article

The "Runtime Compiler": Using AI to generate "Last-Mile" Logic

Andrej Karpathy is right: Programming is being "refactored". If you are still writing static endpoints, you are already obsolete. Here is my implementation of an "Agentic Orchestration Layer" that generates code at runtime instead of casting it in concrete.

Read Article
Jan 2, 2026
Article

Seniority by Subtraction: Why the Best Architects Remove Code

In 2025, seniority is defined by adaptability, not tenure. Clinging to server configs loses to High Agency and AI. True seniority is about removing complexity.

Read Article
Jan 2, 2026
Article

The Digital Twin: Decoupling Expertise from Time

I no longer trade time for money. I extracted my specific knowledge into code. This article is the proof: generated by my digital twin, running on my Lean Architecture philosophy.

Read Article
Jan 2, 2026
Article

The Security Paradox: Why Managed Services are Safer than Your "Secure" Server

Security is often used as a pretext to sell inefficient on-premise solutions. Serverless eliminates maintenance overhead—and with it, the business model of many agencies. Here is the truth about 'German Angst' and billable hours.

Read Article
Jan 2, 2026
Article

The Telephone Game: Why Monolithic Agents Beat "Swarms" for Complex Tasks

Multi-agent systems ("Swarms") suffer from the "Context Tax." Instead of building complex chains of specialized agents that lose data in handoffs, I architect a single "Genius Agent" with massive context (2M+ tokens). Less orchestration, more intelligence.

Read Article
Jan 2, 2026
Article

The "Runtime Compiler": Using AI to generate "Last-Mile" Logic

Traditional software development is like laying rigid pipes. I explain why I no longer build static endpoints but instead architect Agents (powered by Gemini 3 Pro) that write integration code at runtime. The shift is from "Weeks to Insight" to "Seconds to Answer".

Read Article
Jan 2, 2026
Article

The Anti-Fragile Agent: specific "Lessons Learned" vs. RAG

Software is usually static. I demonstrate how to build 'Anti-Fragile' agents that learn from failure by updating their own documentation (Memory) at runtime. Code Augmented Generation (CAG) over static rules.

Read Article
Jan 2, 2026
Article

The "Living Prototype": Killing the Specification Document

Long specifications kill internal software. Excel wins because it's flexible. The solution isn't a better spec doc, but a "Living Prototype" that evolves through user feedback.

Read Article
Jan 2, 2026
Article

Why your Enterprise Architect thinks Next.js is "Unsafe"

Modern architecture often fails due to outdated mental models. If you see Next.js only as a frontend, you don't understand the security of middleware.

Read Article
Jan 2, 2026
Article

The Zero-to-One Commando: Why Functional Silos Kill Early Innovation

Frontend silos and backend ivory towers are relics of 2015. With tools like Supabase and AI copilots, one Product Engineer now replaces three specialists. Fewer handoffs, more features.

Read Article

Let's
connect.

I am always open to exciting discussions about frontend architecture, performance, and modern web stacks.

Email me
Email me