Agentic Engineering

Most people use it to write code. I use it to build organizations.

Agentic Engineering — a term Andrej Karpathy brought into focus in early 2026 — is the emerging discipline of building software systems with AI agents at their core. The industry conversation centers on development methodology: how to architect, orchestrate, and govern autonomous coding agents.

I take it one layer up.

They talk code orchestration. I talk organization design.

My work applies Agentic Engineering to organizational design: how one person, aided by a system of AI agents, can produce the output of an entire organization. Not faster coding. Structural capability.

Three principles

1. Agency over Code — The goal is capability formation, not code output.
2. Architecture over Prompts — Not prompt tricks; structural orchestration.
3. Constitution over Speed — Autonomy must be governed to remain controllable and evolvable.

Frameworks

The following frameworks emerged from building real products, not from theory:

• What Is Agentic Engineering: The full definition, from the organizational angle.
• AOA — Agentic Organizational Architecture: Five layers from strategy to execution.
• OTLT — Two-Time Model: Why capability acceleration without structural coherence leads to failure.
• Capability Density: Engineering output capacity per unit.

Field reports

The frameworks above weren't reasoned into existence. They were extracted from shipping. Each essay below is a forensic look at one decision, one failure mode, or one structural pattern from the field:

• Organization Design Is Agentic Engineering — 4 days, 52 modules, 105,000 lines of code, zero lines written by me. Mintzberg's five coordination mechanisms map cleanly onto five Agentic architectures. The discipline is not new. The substrate is.
• From HR to AI: What 18 Years of Managing People Taught Me About Managing Agents — Managing fifty AI Agents and managing fifty people is the same job. Span of control, escalation paths, performance reviews. The HR playbook ports.
• 95% of People Are Stuck Optimizing Prompts. They Are Wrong. — Prompt optimization is the wrong altitude. Agents aren't tools — they're coworkers. The scarcest skill in this era is people who understand both organizations and Agents.
• Most "AI Agents" Are Just Chat Boxes in a Costume — A seven-rung ladder for separating real autonomy from LLM wrappers. Use it on any "AI agent" pitch in the next five minutes.
• Wake Up: I Wrote My Department in Code — AI isn't an employee. It's an organizational capability module. The stronger the capability, the more it needs a constitution.

Products built this way

Every product below is built by one person using these principles:

• One Page — Focus in one gesture. → Try it
• MARGIN — Natural language → Action. → Try it
• FileFlow — Safe automation. → Try it
• Fairmate — Labor-relation rules, evidence, and risk boundaries in one AI system. → Try it
• SFA — AI sales coach.

If this resonates

I'm not selling consulting. I'm not running a course. I'm building products and writing about what works and what breaks.

If you're sitting on the same intersection — organizational practitioner who started touching AI, or AI builder who started thinking about people and structure — I'd want to hear from you.

→ Find me