X/Twitter Highlights: What Industry Leaders Say About Agentic Engineering

📢 The Agentic Engineering Storm on X/Twitter

In 2025-2026, X/Twitter became the most active platform for AI coding discussions. From Karpathy coining “vibe coding” to proposing “agentic engineering,” from Amodei predicting the end of software engineering as we know it, to CNBC journalists testing AI-built apps — each tweet reshaped our understanding of programming’s future.

Let’s unpack the key voices that are changing the industry.

🧠 Karpathy: The Birth of “Agentic Engineering”

Original Post Breakdown

In late 2025, Andrej Karpathy posted a tweet that sent shockwaves through the tech world:

“There’s a new kind of coding I call ‘agentic engineering’. The human becomes the architect, the reviewer, the taste-maker. The AI agents write ~95% of the code. You guide, review, iterate. It’s not vibe coding — it’s structured, disciplined, and scalable.”

Key Takeaways:

1. Evolution from vibe coding: Karpathy himself coined “vibe coding” in early 2025, but quickly realized that casual coding — while fun — couldn’t support production-grade software
2. Clear human role definition: Architect, reviewer, taste-maker
3. Emphasis on discipline: Not “go with the flow” but “structured, disciplined, and scalable”
4. Quantified contribution ratio: AI writes ~95% of the code, but that 95% is directed and constrained by the human’s 5%

Community Reactions

The tweet garnered tens of thousands of reposts and sparked intense discussion:

Supporters:

• 🟢 “Finally someone gave AI coding a serious name beyond the vibe coding joke.”
• 🟢 “This is exactly what our team has been doing — we just didn’t have a name for it.”
• 🟢 “The ‘architect’ framing is perfect. I’m writing less code than ever, but each line carries more decision weight.”

Skeptics:

• 🔴 “Isn’t this just fancy prompt engineering with a new label?”
• 🔴 “If 95% of code is AI-written, who’s responsible when bugs ship?”
• 🔴 “How do junior engineers grow? If they never write code from scratch, how do they learn architecture?”

💀 Amodei: “Software Engineering Will Be Obsolete in 6-12 Months”

The Controversial Statement

Dario Amodei, CEO of Anthropic, stated in a podcast interview:

“AI will effectively be doing the vast majority of the work of software engineering within 6 to 12 months… The role of a software engineer will change dramatically.”

This blew up on X.

Multiple Perspectives

The Optimists:

• 🟢 “He said ‘role will change,’ not ‘profession will die.’ Just like the shift from manual compilation to automated compilation — programmers didn’t disappear.”
• 🟢 “This perfectly validates agentic engineering: engineers go from ‘writing code’ to ‘directing AI to write code.’”
• 🟢 “Finally we can spend time on what actually matters — architecture design, product thinking, user experience.”

The Pessimists:

• 🔴 “Easy to say from the CEO chair, but what about the entry path for junior engineers?”
• 🔴 “6-12 months is wildly optimistic. AI can’t even handle simple state management reliably.”
• 🔴 “A CEO saying this is just hyping their own products, right?”

The Realists:

• 🟡 “Probably true for simple CRUD apps, but complex distributed systems? Not yet.”
• 🟡 “The question isn’t whether AI can write code — it’s whether AI can understand requirements. It can’t, yet.”

Industry Impact

Despite being provocative, Amodei’s statement accelerated the industry’s attention to AI coding:

• More companies began evaluating AI coding tools
• Engineers started seriously considering their “irreplaceability”
• Educational institutions began adjusting curricula to include AI collaboration

📖 Vibe Coding Shortlisted for Collins Word of the Year

What Happened

In late 2025, Collins Dictionary shortlisted “vibe coding” for its Word of the Year. While it didn’t win the top spot, making the shortlist demonstrated the concept’s mainstream cultural penetration.

Collins Dictionary’s definition:

vibe coding /vaɪb ˈkəʊdɪŋ/ — A style of computer programming in which a person describes what they want in natural language and allows AI to generate the code, often without fully understanding the output.

Reactions on X

Tech Community:

• 🟢 “A programming term entering a mainstream dictionary shows AI coding is no longer a niche toy.”
• 🟡 “Collins defining ‘not understanding the output’ as a core feature — that’s actually vibe coding’s weakness, not a selling point.”
• 🔴 “When a coding methodology is characterized by ‘not understanding the code,’ should we be worried?”

Cultural Significance:

• Vibe coding evolved from a half-joking Twitter concept to a mainstream cultural term
• It reflects how rapidly AI is penetrating everyday language and culture

📺 CNBC’s Vibe Coding SaaS Replication Experiment

What Happened

A CNBC journalist conducted a headline-making experiment: using vibe coding to attempt replicating a well-known SaaS product. The journalist had no programming background and relied entirely on AI tools (primarily Cursor and Claude).

Experiment Results:

• ✅ Successfully built a basic user interface
• ✅ Implemented core CRUD functionality
• ✅ Deployed to the cloud and made it publicly accessible
• ❌ Security concerns (no proper authentication)
• ❌ Performance issues (unoptimized database queries)
• ❌ Poor maintainability (messy code structure)
• ❌ When requirements got complex, AI started making frequent errors

The X Debate

This experiment sparked polarized discussion on X:

“AI Is Everything” Camp:

• 🟢 “A non-programmer built a working SaaS. This is revolution!”
• 🟢 “The barrier to starting a business has never been lower.”
• 🟢 “The first version doesn’t need to be perfect. Get users first. This is how MVPs should be done.”

“AI Isn’t Enough” Camp:

• 🔴 “A SaaS without security? That’s not a product, it’s a security incident.”
• 🔴 “CNBC showed ‘can you build it’ but not ‘can you maintain it.’”
• 🔴 “Wait until this product has 100 users — those unoptimized queries will crash the server.”
• 🔴 “This is exactly why we need agentic engineering instead of vibe coding.”

The Measured Analysis Camp:

• 🟡 “Vibe coding is for validating ideas. Agentic engineering is for building products. They’re not contradictory.”
• 🟡 “The question isn’t whether AI can write code — it’s who takes responsibility for code quality.”
• 🟡 “Best interpretation of this experiment: AI lowered the barrier for prototypes, but NOT for production software.”

🔥 Community Debates & Ongoing Controversies

Debate 1: The Future of Junior Engineers

One of the most heated topics on X:

• 🔴 Worried Camp: “If AI writes 95% of the code, how do junior engineers learn? How do they grow? Who will hire the inexperienced?”
• 🟢 Optimist Camp: “Junior engineers will become ‘junior AI engineering supervisors.’ The entry method changes, but the growth path remains. Just like calculators didn’t eliminate the need to understand math.”
• 🟡 Pragmatic Camp: “In the short term, junior positions will likely shrink. Long term, new roles and opportunities will emerge. The transition period will be painful.”

Debate 2: Code Quality & Security

• 🔴 “AI-generated code looks great but hides more bugs. Because you think you understand it, but you don’t.”
• 🟢 “With rigorous testing and code review, AI code quality is actually very high. The problem is in human processes, not AI itself.”

Debate 3: Creativity vs. Execution

• 🔴 “When everyone uses the same AI to write code, will software become increasingly homogeneous?”
• 🟢 “AI frees up execution-level creativity, letting humans focus on higher-level innovation — product design, user experience, business models.”

Debate 4: Open Source vs. Commercial

• 🔴 “The best AI coding tools cost money. The wealth gap will widen.”
• 🟢 “Aider and Goose are open source. Ollama + local models cost nothing. Tools aren’t the barrier — knowledge is.”

📌 Key Figures & Their Positions

🔮 Our Take

Looking across the X discussions, here’s what we believe:

1. Agentic engineering is the natural evolution of vibe coding — from “fun” to “functional”
2. Software engineering won’t become “obsolete,” but it will “transform” — less code writing, more design and review
3. Tool democratization is happening — but excellent engineers remain scarce
4. The real challenge isn’t technology — it’s how teams adapt to new ways of working

2026 is just getting started. The pace of this transformation may be faster than anyone expects. 💫

Ready to start practicing agentic engineering? Read our Getting Started Guide and Tools Comparison.