Agentic engineering: vibe coding was phase one only?
Agentic engineering didn’t appear just because; it’s the recent sequel to another viral moment – vibe coding. And one that’s already gaining momentum.
Andrej Karpathy, the author, says his original tweet was basically a shower-thought that turned into headlines. And one year later, he posts about what he frames as the next phase.
Software engineers have once been defined by how they wrote, tested, debugged, and refined code strings. Now they are defined by how they direct.
The change began with the term “vibe coding” – a relaxed, intuitive approach of prompting AI-based models. That shifted into spawning “agentic engineering”, the new maturity curve of transformative, AI-backed flows – agentic systems that build, test, debug, and iterate, and that with minimal human keystrokes.
What is the news?
February 4, Andrej Karpathy – who coined “vibe coding” last year – has pushed the conversation even further. He sees the trend as the next phase: “agentic engineering.”
He frames the change as semantic but meaningful:
- “agentic” because the engineer mostly orchestrates the agents and does not type every line
- “engineering” because it’s craft with its own depth
That phrasing has sparked broad coverage and debate about how we should restructure roles and processes.
Let’s get into it.
Who is Andrej Karpathy
Andrej Karpathy was among OpenAI’s founders in 2015 and led Tesla’s Autopilot AI efforts from 2017 to 2022. To date, from 2024 to 2026, he’s building Eureka Labs, a unique education project that aims to revolutionize educational approaches by combining human curriculum with trained “AI teachers.”
He remains an influencer who has gone viral multiple times – he renames a workflow, and the market listens.
What is vibe coding?
Vibe coding is when software developers stop typing code line-by-line and start to “direct” coding assistants. You drop the intent, architecture, constraints, and other high-level nuances, and the model covers the rest – fast, messy, mostly clean.
Vibe coding has shifted the writing and testing to shaping clear prompts and validating the outputs AI suggests. It’s less about syntax and more about direction.
I’ve had a Twitter account for 17 years now (omg) and I still can’t predict my tweet engagement basically at all. This was a shower of thoughts throwaway tweet that I just fired off without thinking but somehow it minted a fitting name at the right moment for something that a lot of people were feeling at the same time, so here we are: vibe coding is now mentioned on my Wikipedia as a major memetic “contribution” and even its article is longer. lol
Andrej Karpathy’s post on agentic engineering – a tweet that became a “contribution”
What is agentic engineering?
Agentic engineering is basically vibe coding but ready for production – it does more than coding augmentation. You manage entire fleets of agents that generate code autonomously, run tests, fix bugs, and report back after – all while you supervise.
Agentic engineering is leveling up complexity: architecture, guardrails, observability, governance – all included. That’s where serious teams start separating from beginners and hobbyists.
The one thing I’d add is that at the time, LLM capability was low enough that you’d mostly use vibe coding for fun throwaway projects, demos and explorations. It was good fun and it almost worked. Today (1 year later), programming via LLM agents is increasingly becoming a default workflow for professionals, except with more oversight and scrutiny. The goal is to claim the leverage from the use of agents but without any compromise on the quality of the software. Many people have tried to come up with a better name for this to differentiate it from vibe coding, personally my current favorite “agentic engineering”
Andrej Karpathy’s post on agentic engineering – a trend to watch
How is agentic engineering different from vibe coding?
Short answer:
- Vibe coding is a creative shortcut
- Agentic engineering, on the other hand, is a repeatable process you can really trust
Diving deeper:
| Vibe coding | Agentic engineering | |
| Objective | Speedy prototyping and experimentation | Reliable delivery and maintenance |
| Execution | – Human prompts – Single-model response | – Human oversight – Agent system |
| Oversight | Human review | Built-in gates |
| Debugging | Manual debugging | Built-in testing and patching |
| Traceability | Weak, hard-to-audit | Strong, auditable |
| Scalability | Best fitting hobby projects | Specifically designed for scale |
| Security posture | Often lax and unpredictable | Policy-driven, automated |
| Skill shift | – Prompt craft – Rapid validation | – Architecture – Guardrails – Observability – Governance |
Human oversight
- Vibe coding is trusting the moment: you prompt, get code, and tweak – that’s it
- Agentic engineering is complex: agents suggest and execute, humans approve
Goal-driven decomposition
- Vibe prompts often ask for an action and accept what comes back after
- Agentic setups will break work down, so each agent has an individual, verifiable deliverable
Code iteration
- Going with vibe coding, you have to cover quality assurance (maybe with some assistance)
- But with agentic engineering, the system will choreograph implementation, test, refactor, and security agents working in cycles
Data governance
- Vibe outputs will live in logs and diffs with very little context
- Agentic systems will attach the spec, test results, approval records, and reasoning
How to “agentic engineer” in practice
Agentic engineering isn’t magic – it’s method.
First architecture, then prompts
Strong projects always begin with defining the specs – architecture, constraints, and everything in between. Your directions will determine future outcomes.
Be specific
Effective prompting means assigning scoped tasks with defined acceptance criteria, not vague feature requests. The tighter the brief, the better the outcomes.
Treat review like shipping is tomorrow
Professional teams will treat agent outputs as such directly related to production, not just experimental drafts. Careful review is critical to ensure sound logic, safe dependencies, and functionality.
Repeat tests until confident
Robust frameworks can turn agent systems from situational, uncertain assistants into powerful project tools. When validation is automated, agent systems can refine their outputs until results will meet defined standards.
Are we already using agentic engineering?
Some teams already implement agentic workflows; other teams are experimenting and adapting their strategy. The divide isn’t tools or capability – it’s maturity.
In practice you’ll see a spectrum:
- Hobbyists & individual programmers: rapid experiments, quick prototypes – more speed, less stability
- Product teams: feature scaffolding, test generation, iterations, debugging
- Startups & engineering-forward groups: agentic systems to handle daily tasks to free more resources
- Top-tier orgs: fully orchestrated multi-agent workflows running alongside CI/CD pipelines
In short, it’s happening, but unevenly.
Pro tip: don’t wait to learn & adapt.
What to expect next
This evolution will facilitate greater productivity – but please don’t panic, it won’t look like coding disappearing. It’ll transform our responsibilities and approaches.
The changes might include:
- Evolving & emerging roles
- Reimagined workflows
- Structural changes within companies
- The teams might shrink in headcount but grow in output
- The bottleneck will move from coding to decision-making
- And new success metrics
The engineer isn’t replaced, he’s promoted from builder to director.
Invest today in people and pilots, measure outcomes, then harden the winners – turn hype into advantage.
How we can help
We design agent workflows, context layers, and guardrails that make AI outputs both reliable and auditable. Not just impressive demos that collapse under workloads.
How about finally moving from experiments to dependable agentic engineering?
Our expertise:
Our services:
FAQ
Vibe coding is a new style where users can describe the idea and let models handle the writing and iteration:
- You guide the intent and steer the process
- The model is covering the routine coding stuff
Agentic engineering, the following maturity curve, is the production-grade evolution:
- You supervise the strategy, the execution, and quality
- The planning, code generation, testing, fixing, and reporting is done by orchestrated agentic systems
Prompt engineering means structuring the inputs to models, so that they produce context-appropriate outputs. It focuses on wording, format, examples, and instructions.
Context engineering means sculpting entire environments to ensure the model’s consistent behavior & output.
That includes memory management, system instructions, tool access, data sources, and requirements.