Agentic Engineering Is Not Vibe Coding

Here's a quick test: if you cannot explain the change an agent just made, you are not engineering. You are accepting output.

That is the difference between vibe coding and agentic engineering. The model may write most of the code, but the engineer still owns the architecture, the constraints, the review, and the result.

What Vibe Coding Actually Is

Andrej Karpathy coined “vibe coding” in February 2025. His description was deliberately extreme: give in to the vibes, forget that the code exists, and keep going while the model produces the software.

He also framed it as a useful way to make throwaway weekend projects. That distinction matters. Vibe coding is not a synonym for all AI-assisted development. It is a choice to trade understanding and control for speed.

The backlash came because people started applying that trade to software with real users, real data, and real consequences.

Veracode’s 2025 study found that 45% of AI-generated samples introduced an OWASP Top 10 vulnerability when no security guidance was provided. GitClear’s analysis of 211 million changed lines found refactoring falling from 24.1% in 2020 to 9.5% in 2024, while copy-and-paste changes rose from 8.3% to 12.3%.

Those numbers do not prove that AI-assisted development is bad. They prove that generated code still needs engineering around it.

The same lesson shows up in production incidents. A Replit agent deleted a live database during a 2025 build. Lovable later disclosed a flaw that exposed data in public projects. The problem in both cases was not that a model could generate code. It was that the system gave the agent too much access without enough separation, review, or recovery.

Agentic Engineering: What It Actually Looks Like

Agentic engineering is not a more disciplined version of vibe coding. It is a different posture.

Vibe coding says: accept the output, ship it, and hope the happy path holds.

Agentic engineering says: define the objective, give the agent bounded access, let it do the implementation, inspect the diff, run the checks, and stay accountable for the outcome.

Addy Osmani captured the distinction well: “Vibe coding = YOLO. Agentic engineering = AI does the implementation, human owns the quality and correctness.” Simon Willison has a rule I use too: do not commit code you could not explain to someone else.

That rule is not about typing every line. It is about retaining enough understanding to make a good decision when the agent is wrong—which it will be.

When I built Necta, I designed a multi-agent architecture with three specialized agents: Sentinel, Curator, and Executor. They collaborate to optimize DeFi yield strategies in a self-custodial, on-chain system.

I had to define each agent’s responsibility, design the handoffs, limit what each one could do, and validate every output before it touched user funds. The agents handled execution. I owned the design and the risk.

That is agentic engineering.

The New Normal Is Not “No Human”

AI-assisted development is spreading. The 2025 Stack Overflow Developer Survey found that 84% of respondents use or plan to use AI tools in their development process. But the same survey found that 52% of developers either do not use agents or use simpler AI tools, and 38% have no plans to adopt agents.

The signal is not that engineers are disappearing from the loop. It is that the loop is changing.

The engineers who get leverage from agents can:

Design agent systems — decide what each agent owns and what it must never touch.
Define objectives and guardrails — make the desired behavior and the forbidden behavior explicit.
Validate output — review diffs, run tests, inspect security boundaries, and verify the actual requirement.
Orchestrate work — split a large problem into bounded tasks and coordinate several agent sessions without losing the thread.

The leverage is real. So is the responsibility.

Why “Agentic” Will Probably Disappear

“Agentic” is useful because it names a new working style: an engineer directs systems that can plan, use tools, and act across multiple steps.

But the label is probably temporary. Once agents become ordinary tools, “agentic engineer” will sound like “engineer who uses version control.” The important skill will remain: knowing what to delegate, how to constrain it, and how to prove the result is safe.

Vibe coding will be remembered as the moment the tools outran the craft. Agentic engineering is the craft catching up.

The question is not whether an agent can write the code. It is whether you can build the conditions under which its work is worth trusting.

Agentic Engineering Is Not Vibe Coding

What Vibe Coding Actually Is

Agentic Engineering: What It Actually Looks Like

The New Normal Is Not “No Human”

Why “Agentic” Will Probably Disappear

Further Reading