AI engineering
Agents, RAG, MCP, structured output, and evals in TypeScript and Python. Built as real systems, not notebook demos.
Production AI engineering for experienced developers
AI engineering that survives production.
I build real AI systems and explain the parts that short tutorials skip: architecture, deployment, observability, evals, and what happens when things break.
What Stevinator covers
The engineering behind useful AI.
One practical video at a time, supported by code, written notes, and public experiments.
Visit the YouTube channelArchitecture and shipping
Deployment, observability, retries, cost, failure modes, and the engineering work between a prototype and a product.
Senior craft and leadership
Practical judgment from software delivery and project management: teams, trade-offs, deadlines, and how AI changes the work.
Public experiments
Same problem. Different models. Evidence over claims.
I give multiple AI models the same constrained task, preserve the first result, and compare what they actually produce.
Browse all experimentsFeatured comparison
Pelican On A Bicycle
A visual gotcha: can the model draw a recognizable pelican riding a bicycle?
Open the comparisonAnonymous output A
Anonymous output B
Writing
Notes for people who ship software.
Written companions to the videos, plus practical notes on programming, architecture, project management, and engineering decisions.
Browse all writingI Built an AI Development Team in My Terminal
Projectinator turns one app idea into a Scrum backlog, working files, and a browser-tested result. Here is how the pipeline works, what it costs, and where it still needs human judgment.
Read the companion →Video planned
Keep Learning to Code: AI Changes the Job, Not the Need for Programmers
AI can write more code every month. That is not a reason to stop learning programming. It is a reason to become the developer who can direct, verify, and deliver with it.Why I Use AI Coding Agents in the Terminal
I use Codex and Claude Code only from the terminal. Here is the exact repository workflow, prompt structure, and safety boundary you can copy.The complete Exercism solution archive remains available separately from the main editorial feed.
Open the Exercism archiveAlso shipping
Software is part of the practice.
The apps are where product decisions, implementation details, and real-user constraints meet.
View all apps
About Stepan
The experience behind the experiments.
I'm Stepan Manookian. Friends call me Steve. I have spent more than twelve years across full-stack development, IT project management, automotive software, and independent product work.
Stevinator is where I build production-minded AI systems in public, explain the decisions, and share what changes when software meets real users, teams, and deadlines.