What is Agentic Engineering?

What is Agentic Engineering? And why it's not "vibe coding" with a fancier name.

I haven't written a for loop in months. That's not a flex — it's a data point. I'm a Principal Software Engineer at Minted with over 20 years of experience, and the way I write software has fundamentally changed in the last six months. Not because I forgot how to code, but because I stopped being the one who types it. Let me explain.

From Autocomplete to Orchestration

My journey started simply enough. In late 2024, I picked up GitHub Copilot and saw immediate productivity gains from its code completion. By mid-2025, I'd moved to Cursor's Agent Mode and the gains compounded. But the real shift happened in November 2025, when I started using Claude Code with Anthropic's most capable model.

I had a concrete problem: modernizing legacy applications that are already end-of-life (EOL), converting old PHP code to modern Python, and migrating them to our modern microservices stack. The kind of work that's tedious, high-stakes, and usually takes quarters to complete.

The code Claude Code generated was good. Not "good for AI" — genuinely good. And when I layered in proper planning, context engineering, and tight feedback loops, the quality and velocity became something I'd never seen in two decades of shipping software.

That's when it clicked: I wasn't coding anymore. I was engineering — just with a fundamentally different workflow.

So What Is Agentic Engineering?

The term was coined by Andrej Karpathy in early 2026, and it describes exactly what I was experiencing: the transition from writing code to orchestrating AI agents that write code for you.

Think of it this way. You're no longer an individual contributor typing out syntax. You're an engineering manager whose direct reports are highly capable, incredibly fast — and sometimes dangerously naive — AI agents.

Your job shifts to the things that actually matter:

Architecture & Orchestration. You provide the high-level direction and system design. You decompose complex systems into subtasks that AI agents can tackle iteratively. You're the one who knows what to build and why.

Validation & Taste. AI can write the code, but it doesn't know what "good" looks like for your specific codebase, your business constraints, or your team's conventions. You provide the taste — enforcing patterns, reviewing agent-generated PRs, and holding the quality bar.

Human-in-the-Loop Oversight. You don't hit "generate" and deploy. You steer agents when they drift, validate their tests, and secure their output. You are fundamentally in the loop at every step.

Managing Technical Debt. Your primary focus shifts to preventing AI slop — governing agents so they build maintainable software instead of fragile scripts that happen to pass a demo.

This Is Not Vibe Coding

I need to be direct about this, because LinkedIn is drowning in posts about "vibe coding" — the idea that you can describe a vibe in natural language and an AI will build you a million-dollar SaaS while you sip cold brew.

My eyes roll every single time.

Here's the reality: vibe coding works beautifully on the happy path. Weather apps. To-do lists. Tutorial projects. But software engineering is 80% edge cases, error handling, and state management. The moment your vibe-coded app hits an unexpected API timeout or weird user input, the vibe crashes.

Worse: if you don't understand the code the AI generated — if you can't read it, debug it, reason about its failure modes — you don't own a product. You own a liability.

Agentic Engineering is the professionalization of AI-assisted development. It brings engineering rigor back into the picture. It acknowledges that while AI models are now capable enough to generate the logic, they still require a seasoned engineer to make the output production-ready, scalable, and secure.

What AI Actually Is (From Someone Who Works With It Daily)

After months of working side-by-side with AI agents, here's my honest assessment:

AI writes plausible code, not necessarily correct code. It's a prediction engine that generates solutions based on training data. Plausible code can harbor silent bugs, race conditions, and security vulnerabilities that look perfectly fine in a PR diff.

AI has zero business context. It doesn't know why the marketing team insists on that specific user flow, or why the legacy database needs that bizarre workaround. Give it a pristine, isolated problem and it'll nail it. Drop it into a 10-year-old enterprise codebase with undocumented dependencies and it'll hallucinate wildly.

The best mental model: AI is a hyper-competent junior developer who has memorized every piece of documentation on the internet but has zero real-world judgment. You still review its work. Line by line.

Where This Is Heading

At Minted, we've been using AI coding assistants for over a year. We're now ready for the next level — Agentic Software Engineering at scale. I'm leading the effort to equip our engineering teams with the knowledge, tooling, and infrastructure to grow this practice across the organization.

We're not alone in this. Stripe recently published a two-part blog series on their internal "Minions" system — it's one of the most pragmatic and grounded looks at agentic engineering I've seen. No hype, just a serious engineering organization showing how they've made AI agents work within real-world constraints. That's the kind of approach we're striving toward.

The Bottom Line

Will AI replace developers? No.

Developers who learn to orchestrate AI will replace developers who refuse to use it.

The job title might stay the same, but the job description is shifting fast. We're moving from the era of syntax typists to system architects. The work is no longer memorizing React hook arguments — it's system design, context engineering, rigorous review, and translating ambiguous human needs into precise technical constraints.

When someone on LinkedIn posts "AI built my entire app," what they usually mean is: "I spent 40 hours aggressively debugging and coaxing an AI to build a relatively simple wrapper around an existing API."

That's still development. It just uses a different keyboard.

And honestly? The engineers who've spent years building judgment — who understand distributed systems, failure modes, and the unglamorous reality of production software — are the ones best positioned for this shift.

Agentic Engineering doesn't make senior engineers obsolete. It makes your experience more valuable than ever.

I'm Jon Cortez, a Principal Software Engineer at Minted . I write about the intersection of AI and software engineering — the real stuff, not the hype. Follow along if you're interested in what Agentic Engineering actually looks like in practice.