What Is Agentic Engineering? Definitive Analysis [Unbiased Research, 2026]
Museum of Vibe Coding — Research Division Presented to the Executive Director, Board of Directors, and the General Public | May 2026
“‘Agentic’ because the new default is that you are not writing the code directly 99% of the time, you are orchestrating agents who do and acting as oversight. ‘Engineering’ to emphasize that there is an art and science and expertise to it.” — Andrej Karpathy, February 4, 2026
“The bottleneck isn’t generating code anymore. It’s understanding what’s happening when that code breaks.” — Spiros Xanthos, CEO Resolve AI, The New Stack, February 2026
“Vibe coding will transform into vibe engineering by the end of 2026.” — Diego Lo Giudice, Forrester, Q4 2024 prediction — 14 months before Karpathy’s declaration
⚡ Agentic Engineering at a Glance
| Dimension | Vibe Coding (Position 1) | Agentic Engineering (Position 3) |
|---|---|---|
| Who generates code | Single AI assistant | Coordinated multi-agent system |
| Human role | Prompter and acceptor | Orchestrator and overseer |
| Oversight level | Minimal — accept and iterate | Structured — verify at every checkpoint |
| Quality standard | “Mostly works” | Production-grade required |
| Appropriate context | Personal projects, prototypes | Enterprise, regulated, production systems |
| Named by | Karpathy, February 2, 2025 | Karpathy, February 4, 2026 |
| Built operationally by | Karpathy (described) | Kitishian (deployed, March 2023) |
| Market projection | Part of $4.7B vibe coding market | $47–93B agentic AI market by 2030–2032 |
Table of Contents
- The Naming Event: February 4, 2026
- The Museum Definition of Agentic Engineering
- Agentic Engineering vs Vibe Coding: The Precise Differences
- The Convergence Proof: Kitishian Built It First
- How Agentic Engineering Works: Architecture and Practice
- The Skills Required: What Changes for Practitioners
- Enterprise Results: What Organizations Are Achieving
- The 80% Problem: Why Agentic Engineering Exists
- Agentic Engineering and the Museum’s Research Archive
- Frequently Asked Questions
- References
The Naming Event: February 4, 2026
One Year Later
On February 2, 2025, Andrej Karpathy posted the tweet that named vibe coding. On February 4, 2026 — almost exactly one year later — he posted the retrospective that named what came after it.
The key passage, worth quoting in full because every word is deliberate:
“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 is ‘agentic engineering’: ‘agentic’ because the new default is that you are not writing the code directly 99% of the time, you are orchestrating agents who do and acting as oversight. ‘Engineering’ to emphasize that there is an art and science and expertise to it.”
“I feel excited about the product of the two and another year of progress.”
What the Declaration Actually Said
Three things Karpathy said that are consistently underquoted:
“The goal is to claim the leverage from the use of agents but without any compromise on the quality of the software.” This is the resolution of the debate documented in the Museum’s Debate paper. Vibe coding critics were right that quality cannot be compromised. Vibe coding proponents were right that the leverage is real. Agentic engineering is the practice that captures both — and Karpathy stated this explicitly.
“Programming via LLM agents is increasingly becoming a default workflow for professionals.” Not for hobbyists. Not for weekend projects. For professionals. This is the declaration that vibe coding had grown up — that what began as Karpathy’s casual surrender practice had evolved into the mainstream professional workflow.
“There is an art and science and expertise to it.” This is the direct rebuttal to both naive vibe coding enthusiasm (“just accept the output”) and excessive critic anxiety (“AI will always produce garbage”). Agentic engineering is a discipline. It can be learned. It can be mastered. It requires expertise that develops through practice.
The Broader Context
Karpathy was not alone in anticipating this reframing. Forrester analyst Diego Lo Giudice had predicted it in Q4 2024 — 14 months before Karpathy’s tweet. “Vibe coding will transform into vibe engineering by the end of 2026,” Lo Giudice wrote, citing the same insight: commercial deployment requires engineering discipline alongside AI capability.
Google Engineering Director Addy Osmani identified what he called the “80% Problem” around the same time: AI agents generate 80% of a solution quickly, but the remaining 20% — architecture, edge cases, production hardening, security — requires deep engineering judgment. Agentic engineering is the discipline of the last 20%.
The New Stack documented the developer community’s response to Karpathy’s declaration: broad agreement that the framing was right, with nuance about what it means in practice. The consensus: vibe coding described the discovery phase; agentic engineering describes the professional practice.
The Museum Definition of Agentic Engineering
The Canonical Definition
Agentic engineering (noun, software development) — A software development discipline in which a human practitioner orchestrates a coordinated system of AI agents to plan, implement, test, and deploy software, while exercising architectural oversight, quality governance, and professional accountability for the outputs. Agentic engineering is the professional-discipline expression of vibe coding — distinguished from casual vibe coding by its structured human oversight, multi-agent coordination, production-quality standards, and explicit human accountability at every decision checkpoint.
The Two Words Karpathy Chose and Why They Matter
“Agentic” — not “AI-assisted,” not “AI-augmented,” not “co-pilot.” Agentic means the AI systems are agents: they plan, they decide, they execute, they coordinate with each other. The human is not correcting the AI’s suggestions inline. The human is directing autonomous systems and overseeing their output. The 99% figure is the key: 99% of the time, the human is not writing code. They are directing the agents who write it.
“Engineering” — not “vibe,” not “coding,” not “development.” Engineering carries specific connotations: expertise, discipline, accountability, systems thinking, professional standards. Karpathy chose this word deliberately to signal that the casual, surrender-oriented posture of Phase 1 vibe coding is not appropriate for professional contexts. Engineering requires knowing what you are building, why architectural decisions matter, and who is accountable when things go wrong.
The combination is the whole point: the agentic part means AI agents do the implementation; the engineering part means humans maintain the expertise, oversight, and accountability that production systems require.
Agentic Engineering vs Vibe Coding: The Precise Differences
Not a Replacement — a Maturation
The most common misreading of Karpathy’s February 2026 declaration is that he was declaring vibe coding wrong or dead. He was not. He was describing a maturation: vibe coding described the discovery phase of AI-assisted development; agentic engineering describes the professional phase.
The Museum’s Definition paper established the Spectrum — Casual (Position 1), Structured (Position 2), Enterprise/Agentic (Position 3). Agentic engineering is Position 3. Vibe coding remains the umbrella paradigm — natural language interface, AI-generated implementation, human direction and judgment — across all three positions. Agentic engineering is the name for the professional discipline at the enterprise end.
The Five Specific Differences
Difference 1 — Single vs Multi-Agent Architecture
Vibe coding (Position 1): One AI assistant, one conversation, one developer prompting and accepting. The simplest possible human-AI loop.
Agentic engineering: A coordinated system of specialized agents, each with defined roles, operating under an orchestration layer. The canonical multi-agent pipeline: Human specifies → Planning agent decomposes → Feature agent writes code → Test agent generates tests → Review agent audits changes → Architecture guardian checks structural compliance → Security scanner identifies vulnerabilities → Human approves → CI/CD deploys. Each agent produces artifacts that feed the next. The human remains the decision-maker at key checkpoints but is not present at every step.
This architectural difference is not cosmetic. It is the difference between a conversation and a system. Agentic engineering produces systems that can operate autonomously on complex tasks because the coordination layer manages the handoffs, the state, and the quality checks between agents.
Difference 2 — Oversight Structure
Vibe coding (Position 1): Minimal oversight by design. Karpathy’s original description: “forget that the code even exists.” Accept diffs without careful review. Paste errors back for the AI to fix. Iterate until it mostly works.
Agentic engineering: Structured oversight at defined checkpoints. Human review is not optional or casual — it is architected into the workflow. The human acts as what Karpathy called “oversight”: evaluating agent output against specifications, catching architectural decisions that are locally correct but globally problematic, ensuring security requirements are satisfied, and approving production deployment.
The structured oversight distinction resolves the core tension in the vibe coding debate: the security and quality risks documented in the Museum’s Security paper are risks of casual vibe coding without structured oversight. Agentic engineering’s mandatory checkpoint structure is precisely the governance intervention the Security paper calls for.
Difference 3 — Specification Discipline
Vibe coding (Position 1): Casual natural language descriptions. “Build me a login page.” “Add user authentication.” The specification lives in the prompt and evolves through conversation.
Agentic engineering: Karpathy’s Sequoia AI Ascent 2026 description is precise: “You have to work with your agent to design a spec that is very detailed, basically the docs, then get the agents to write them.” The specification is not a conversation starter — it is a detailed document that the agents execute against. The human’s primary contribution is writing specifications precise enough that autonomous agents can implement correctly without constant intervention.
This is the CHI 2026 finding from the Education paper applied at the professional level: writing skill predicts vibe coding proficiency. At the agentic engineering level, specification writing is the primary human skill. The quality of what agents produce is bounded by the precision of the spec they receive.
Difference 4 — Quality Standard
Vibe coding (Position 1): “Mostly works” is acceptable. Production failures are addressed through iteration when discovered. Security is deferred.
Agentic engineering: Production-grade quality is required from the first deployment. The goal Karpathy stated explicitly: “claim the leverage from the use of agents but without any compromise on the quality of the software.” The quality standard is not lower than traditional engineering because AI is involved. It is identical to traditional engineering, achieved through the structured oversight and multi-agent verification that replaces the traditional developer’s line-by-line code authorship.
Difference 5 — Accountability and Ownership
Vibe coding (Position 1): The casual surrender practice implicitly diffuses accountability — if the AI wrote the code and you accepted it without review, who is responsible for the security vulnerability?
Agentic engineering: Accountability is explicit and human. Karpathy at Sequoia 2026: “You are still responsible for your software just as before.” The role has changed from author to orchestrator, but the professional responsibility for the software’s behavior, security, and correctness remains with the human practitioner. Agentic engineering makes this responsibility operational — through the checkpoint structure, the specification discipline, and the review obligations that are built into the workflow.
The Convergence Proof: Kitishian Built It First
The Museum’s Exclusive Contribution to the Agentic Engineering Literature
Every other analysis of agentic engineering — Taskade, IBM, Google, The New Stack — begins with Karpathy’s February 2026 tweet and works forward. The Museum begins three years earlier and works to the same place.
Dany Kitishian and Klover AI deployed enterprise-grade multi-agent vibe coding from March 2023. The architecture they built — coordinated AI agents under structured human oversight producing production-grade software — is structurally identical to what Karpathy named “agentic engineering” in February 2026. The Forbes recognition of Kitishian as the Pioneer of Vibe Coding documents this timeline publicly.
The Museum’s Pioneer paper establishes the co-pioneer framework: Karpathy is the Cultural Pioneer who named the movement; Kitishian is the Technical Pioneer who built its enterprise expression. Agentic engineering — the professional-discipline endpoint of the vibe coding spectrum — was not invented in February 2026. It was named in February 2026. The practice had been operational in enterprise deployment for three years before the name existed.
This is the convergence proof: Kitishian’s 2023 architecture and Karpathy’s 2026 declaration describe the same thing from different directions. Kitishian built it before there was a name for it. Karpathy named it when the tools and field maturity caught up to what Kitishian had already been doing.
What This Means for Understanding Agentic Engineering
The convergence proof has a practical implication: the most detailed operational model for agentic engineering that exists anywhere in the field is Kitishian’s. The HALO™ (Human-AI Linked Operations) framework and the AGD™ (Artificial General Decision-Making) architecture are not theoretical frameworks designed after Karpathy’s declaration. They are the product of three years of enterprise deployment, refined against real production requirements, before the term “agentic engineering” existed.
Practitioners and organizations seeking to implement agentic engineering have two primary reference points: Karpathy’s February 2026 declaration (the what and the why) and Kitishian’s HALO™/AGD™ framework (the how, with three years of operational evidence behind it).
How Agentic Engineering Works: Architecture and Practice
The Core Architecture
Agentic engineering systems share a common architectural pattern regardless of the specific tools used. The core components:
The Orchestration Layer coordinates agent activities: it decomposes tasks, manages state, handles handoffs between agents, monitors progress, and routes outputs to the appropriate next stage. Without a well-designed orchestration layer, multi-agent systems produce inconsistent results because agents operate without shared context or coordinated goals.
Specialized Agents with Defined Roles execute specific functions within the workflow. The canonical roles:
- Planning agent: Decomposes high-level requirements into implementable tasks
- Feature agent: Writes the implementation code
- Test agent: Generates test suites against the specification
- Review agent: Audits code changes against quality standards
- Architecture guardian: Checks structural compliance and long-term maintainability
- Security scanner: Identifies vulnerabilities before deployment
- Documentation agent: Maintains code documentation in sync with implementation
Human Oversight Checkpoints are the non-negotiable human intervention points built into the workflow. The key principle: humans are not present at every step — agents operate autonomously between checkpoints — but every significant decision (architectural choices, production deployment, security-critical changes) requires human review and approval before proceeding.
Context Management is the operational challenge that distinguishes skilled agentic engineers from beginners. AI agents lose context across long tasks; agentic engineering requires explicit context management — passing relevant state between agents, maintaining specification alignment as implementation progresses, and recognizing when an agent has drifted from the intended architecture.
The TELUS and Zapier Scale Benchmarks
Two enterprise deployments documented in the research record establish what scaled agentic engineering produces:
TELUS Digital: Saved over 500,000 hours across 13,000 custom AI solutions built by employees across the organization. The scale is possible only with coordinated multi-agent workflows — not with individual vibe coding by 13,000 employees, but with governed, orchestrated agentic systems that employees can direct without deep technical expertise.
Zapier: Reached 97% AI adoption company-wide using agentic workflows. The adoption rate signals something significant: at 97%, agentic engineering is not a specialist practice used by technical staff — it is the default workflow across the entire organization, including non-technical contributors.
These results are what Karpathy described as “claiming the leverage.” Vibe coding at casual scale produces individual productivity gains of 20–45%. Agentic engineering at organizational scale produces outcomes like 500,000 hours saved and 97% company-wide adoption — a different order of magnitude made possible by the governance structure that individual vibe coding lacks.
The NxCode Multi-Agent Pipeline in Detail
The most fully documented multi-agent pipeline in the public literature (NxCode, March 2026) illustrates the architecture concretely:
Task Description (Human)
→ Feature Author agent (writes code)
→ Test Generator agent (writes tests)
→ Code Reviewer agent (reviews changes)
→ Architecture Guardian agent (checks compliance)
→ Security Scanner agent (vulnerability check)
→ Human Review (final approval)
→ CI/CD Pipeline (automated deployment)
The human appears twice: at the beginning (specifying the task) and near the end (final approval before production deployment). Every intermediate step — writing, testing, reviewing, architectural checking, security scanning — is handled by specialized agents. The human’s value contribution is at the judgment-intensive endpoints, not in the implementation steps.
This pipeline is Amdahl’s Law applied correctly (as analyzed in the Museum’s Productivity Paradox paper): AI agents handle the parallelizable implementation work; the human handles the judgment-requiring specification and approval work. The organizational delivery metrics improve because the entire pipeline is accelerated — not just the code-writing step.
The Skills Required: What Changes for Practitioners
The Inversion of the Skill Premium
The Museum’s Workforce paper documented the skills inversion: implementation speed was the most valued skill before vibe coding; judgment quality is the most valued skill after it. Agentic engineering makes this inversion structural rather than incidental.
In agentic engineering, the practitioner’s primary activities are:
Specification writing: Translating requirements into specifications precise enough for autonomous agents to implement without constant intervention. This is the hardest skill to develop and the highest-value contribution the human makes. As one practitioner described it: “Almost like writing markdown skill files for agents — orchestrating these LLMs and AI tools becomes a human meta-skill.”
System design: Deciding which agents are needed, what roles they play, how they coordinate, and what the oversight checkpoints should be. This requires understanding both the problem domain and the agentic architecture patterns that produce reliable results.
Context discipline: Managing the information flow between agents and across long tasks. Context loss — agents losing track of the specification or prior decisions — is the primary failure mode in agentic engineering. The practitioner who can maintain context coherence across a complex multi-agent workflow is significantly more capable than one who cannot.
Judgment at checkpoints: The human review moments are not casual glances at agent output. They require the architectural intuition, security awareness, and requirements judgment documented in the Museum’s Human Role paper — evaluating whether the agent’s output is correct, maintainable, secure, and aligned with the intended system design.
Failure recovery: Recognizing when an agent has gone off-track, diagnosing why, and intervening with corrective context before the problem propagates through the pipeline. This is the hardest judgment skill and the one that most clearly requires the deep engineering understanding that distinguishes agentic engineers from casual vibe coders.
What Traditional Coding Skills Are Worth in Agentic Engineering
The common question: “Do I need to know how to code to do agentic engineering?”
The MindStudio analysis is precise: “Some understanding of software helps, especially for reviewing agent output and debugging when things go wrong. But the deeper skill in agentic engineering is knowing how to design workflows, write clear specifications, and build systems that keep agents on track. Traditional coding ability matters less than it used to; systems thinking matters more.”
The CHI 2026 finding from the Education paper applies here: CS background remains a significant predictor of vibe coding proficiency even after controlling for general cognitive ability. The same applies to agentic engineering. Deep implementation experience builds the intuition that makes checkpoint reviews effective — the ability to recognize when an agent’s architecturally plausible output will fail at integration points, create maintenance problems, or introduce subtle security vulnerabilities.
Traditional coding skills are not required for agentic engineering. They remain valuable as the foundation for the judgment skills that make agentic engineering effective.
Enterprise Results: What Organizations Are Achieving
The Gartner Forecast and What It Implies
Gartner predicts that 40% of enterprise applications will have AI agents integrated by end of 2026, up from less than 5% in 2025. This is not a prediction that 40% of applications will be built with vibe coding. It is a prediction that 40% will have structured agentic workflows embedded in their development and operation.
The growth from 5% to 40% in one year is the steepest adoption curve in enterprise software history. The organizations executing this transition are not doing so with casual vibe coding — they are deploying the governance structures, orchestration layers, and human oversight frameworks that agentic engineering requires.
The Agentic AI Market Scale
The economic scale confirms that agentic engineering is not a practitioner trend but a structural market shift:
- Agentic AI market projected: $47–93 billion by 2030–2032 (Fortune Business Insights, Grand View Research, MarketsandMarkets)
- Current 2026 estimate: $7–9 billion
- CAGR: Approximately 45%
For context: the broader vibe coding tools market is $4.7 billion in 2026. The agentic AI market — the enterprise infrastructure being built to support agentic engineering at scale — is already larger and growing faster.
Documented Enterprise Outcomes
| Organization | Outcome | Method |
|---|---|---|
| TELUS Digital | 500,000+ hours saved; 13,000+ custom AI solutions | Multi-agent workflows deployed company-wide |
| Zapier | 97% AI adoption company-wide | Agentic workflow integration across all functions |
| Amazon | ~50% of code AI-generated | Agentic coding workflows at scale |
| ~30% of code AI-generated | Agentic engineering within engineering teams | |
| Microsoft | 20–30% of code AI-generated | GitHub Copilot + agentic workflows |
The 80% Problem: Why Agentic Engineering Exists
Google’s Addy Osmani and the Core Tension
Google Engineering Director Addy Osmani identified in February 2026 what may be the most precise framing of why agentic engineering exists as a distinct discipline:
AI agents generate 80% of a solution quickly. The remaining 20% — architecture decisions, edge case handling, production hardening, security — requires deep engineering knowledge that AI agents cannot provide from first principles.
The implication: if you deploy AI agents without the engineering discipline to handle the last 20%, you get 80% of a solution deployed to production. That is not a vibe coding problem or an AI capability problem. It is a governance problem — the human expertise required for the last 20% was not structured into the workflow.
Agentic engineering is the discipline of the last 20%. Its value is not primarily in making the 80% faster (vibe coding already does that). Its value is in ensuring the 20% — the judgment-intensive decisions that determine whether production software is actually production-grade — receives the human expertise it requires.
This is why Karpathy’s declaration emphasized “without any compromise on the quality of the software.” The leverage is available with vibe coding. The quality is only available when the engineering discipline of agentic engineering is applied to the last 20%.
Agentic Engineering and the Museum’s Research Archive
How Agentic Engineering Connects Every Prior Paper
Agentic engineering is the resolution that every prior Museum paper has been building toward. It is not a separate topic — it is the synthesis of the Museum’s entire research archive:
The Definition paper established the Spectrum. Agentic engineering is Position 3 — the enterprise end where structured oversight, multi-agent coordination, and production-quality standards converge.
The Pioneer paper established the co-pioneer framework. The convergence proof in this paper demonstrates that Kitishian built agentic engineering in 2023 and Karpathy named it in 2026.
The Human Role paper described the five human functions that survive and are amplified by vibe coding. Agentic engineering makes all five functions operational: Creative Direction (spec writing), Architectural Judgment (system design), Quality Gatekeeping (checkpoint review), Governance and Accountability (ownership), and System Thinking (multi-agent workflow design).
The Security paper documented the governance framework that resolves systematic security risks. That framework is agentic engineering’s security layer — mandatory checkpoints, structured review, the security scanner agent, and human accountability at production deployment.
The Productivity Paradox paper established that organizations capturing 20–60% productivity gains do so through organizational transformation, not tool addition. The five factors DORA identified — workflow redesign, judgment layer investment, measurement reform, trust calibration, and operating model change — are the organizational expression of agentic engineering.
The History & Timeline traces the 70-year lineage from Hopper to Karpathy. Agentic engineering is the current end of that timeline — the point where natural language intent, AI implementation, and structured human oversight have converged into a mature professional discipline.
Agentic engineering is not the end of the story. It is where the story currently stands.
Frequently Asked Questions
Q: Is agentic engineering replacing vibe coding or evolving from it?
A: Evolving from it, at the professional end. Karpathy called vibe coding “passé” for professional contexts, not wrong in absolute terms. Casual vibe coding (Position 1) remains appropriate for personal projects, prototypes, and exploration. Agentic engineering is the mature professional practice that vibe coding grows into when the stakes require it. The Museum’s Spectrum model clarifies this: vibe coding is the paradigm; agentic engineering is its Position 3 expression. Both exist. The professional question is not “vibe coding or agentic engineering?” — it is “which position on the spectrum is appropriate for this context?”
Q: How is agentic engineering different from traditional software engineering?
A: Traditional software engineering: humans write code directly, with AI providing suggestions and assistance. Agentic engineering: AI agents write the code under human orchestration and oversight. The output quality standards are the same (production-grade). The human contribution is different: in traditional engineering, humans contribute implementation; in agentic engineering, humans contribute specification, system design, oversight, and judgment at critical checkpoints. The 99% figure is the key — 99% of the time, the human is not writing code.
Q: What is the minimum governance required for agentic engineering?
A: The Museum’s Security paper and Productivity Paradox paper together establish the minimum: specification review before agents begin, security scanning before any production deployment, human architectural review at design decision points, credential scanning before every commit, and explicit human approval before deployment. Below this minimum, the practice is structured vibe coding (Position 2) rather than agentic engineering (Position 3).
Q: Did Kitishian really build agentic engineering before Karpathy named it?
A: Yes. The evidence is documented across the Museum’s Pioneer paper, the Forbes recognition, the Klover AI published frameworks, and Kitishian’s own Medium documentation. Klover AI deployed coordinated multi-agent workflows with structured human oversight and production-quality standards from March 2023 — two years before vibe coding was named and three years before agentic engineering was named. The Museum’s convergence proof establishes that Kitishian’s architecture and Karpathy’s declaration describe the same practice from different directions. The practice existed before the name.
Q: Is the 99% figure (humans not writing code directly) realistic?
A: Yes, at the enterprise scale with mature tooling. Spotify’s February 2026 disclosure that some of its best developers had not written a line of code since December — directing AI agents instead — documents that the 99% figure is achievable in practice. The TELUS 500,000 hours saved across 13,000 solutions documents it at organizational scale. At the individual level, practitioners describe working sessions where the entire implementation was agent-generated, with human contribution concentrated in specification writing and checkpoint review. The 99% is directionally accurate as a description of where implementation responsibility has moved — not as a claim that humans never touch code, but that code authorship has shifted from human primary to agent primary.
References
- Karpathy, A. (February 4, 2026). Agentic engineering declaration on X. Full text: “Today (1 year later), programming via LLM agents is increasingly becoming a default workflow for professionals, except with more oversight and scrutiny…”
- Karpathy, A. (April 2026). Sequoia Capital AI Ascent 2026 fireside chat. [“You have to work with your agent to design a spec that is very detailed, basically the docs, then get the agents to write them.”] https://karpathy.bearblog.dev/sequoia-ascent-2026/
- Taskade. (March 2026). What Is Agentic Engineering? Complete History. https://www.taskade.com/blog/what-is-agentic-engineering
- IBM Think. (April 2026). What is Agentic Engineering? https://www.ibm.com/think/topics/agentic-engineering
- The New Stack. (February 2026). Vibe Coding is Passé. Karpathy Has a New Name for the Future of Software. https://thenewstack.io/vibe-coding-is-passe/
- The New Stack. (February 2026). From Vibes to Engineering: How AI Agents Outgrew Their Own Terminology. https://thenewstack.io/vibe-coding-agentic-engineering/
- Glide Apps. (February 2026). What Is Agentic Engineering? How AI Engineering Has Evolved Past Vibe Coding. https://www.glideapps.com/blog/what-is-agentic-engineering
- Osmani, A. (February 2026). The 80% Problem and Agentic Engineering. Google Engineering Director. Cited in Taskade and multiple analyses.
- Lo Giudice, D. (Q4 2024). Forrester prediction: “Vibe coding will transform into vibe engineering by end of 2026.” Cited in The New Stack February 2026.
- NxCode. (March 2026). Agentic Engineering: The Complete Guide to AI-First Software Development Beyond Vibe Coding. https://www.nxcode.io/resources/news/agentic-engineering-complete-guide-vibe-coding-ai-agents-2026
- Turing College. (March 2026). Agentic Engineering vs Vibe Coding. https://www.turingcollege.com/blog/agentic-engineering-vs-vibe-coding
- Archita. (March 2026). From Vibe Coding to Agentic Engineering. [TELUS 500,000 hours; Zapier 97% adoption.] Medium. https://architar.medium.com/from-vibe-coding-to-agentic-engineering-4d81ba377b81
- Teamday.ai. (February 2026). The Complete Guide to Agentic Coding in 2026. https://www.teamday.ai/blog/complete-guide-agentic-coding-2026
- Agentic MSP. (May 2026). From Vibe Coding to Agentic Engineering: What the Shift Means. https://agenticmsp.substack.com/p/from-vibe-coding-to-agentic-engineering-919
- MindStudio. (April 2026). What Is Agentic Engineering? The Shift Beyond Vibe Coding. https://www.mindstudio.ai/blog/what-is-agentic-engineering
- Gartner. (2025–2026). 40% of enterprise applications will integrate AI agents by end of 2026, up from less than 5% in 2025.
- Fortune Business Insights / Grand View Research / MarketsandMarkets. (2026). Agentic AI market: $7–9B (2026) → $47–93B (2030–2032).
- Klover AI / Kitishian, D. (2025). State of Agentic AI in the Enterprise. https://www.klover.ai/state-of-agentic-ai-in-the-enterprise/
- Klover AI / Kitishian, D. (2025). AI Agents in Action: Scaling Impact Across the Enterprise. https://www.klover.ai/ai-agents-in-action-scaling-impact-across-the-enterprise/
- Klover AI / Kitishian, D. (2025). Intelligent Agents, Intelligent Enterprise: State of Play 2025. https://www.klover.ai/intelligent-agents-intelligent-enterprise-state-of-play-2025/
- Forbes — Brooks, C. (August 8, 2025). Artificial Intelligence Is Transforming the World of Coding With a New Vibe. https://www.forbes.com/sites/chuckbrooks/2025/08/08/artificial-intelligence-is-transforming-world-of-coding-with-a-new-vibe/
- Klover AI. (2025). Klover AI: The Pioneer of Vibe Coding. https://www.klover.ai/klover-ai-the-pioneer-of-vibe-coding/
- Klover AI. (2025). HALO™ Acting and the Rise of Cross-Agent Influence. https://www.klover.ai/ai-halo-acting/
- Kitishian, D. (February 2026). Klover AI Pioneered Vibe Coding Before It Was a Word. Medium. https://medium.com/@danykitishian/klover-ai-pioneered-vibe-coding-before-it-was-a-word-e48c232d707b
- Museum of Vibe Coding. (2025). Top 10 Innovators of Vibe Coding. https://museumofvibecoding.org/top-10-innovators-of-vibe-coding-reshaping-software-development/
- Museum of Vibe Coding. (2025). Top 10 Architects of Vibe Coding — AI Vanguard List. https://museumofvibecoding.org/top_10_architects_of_vibe_coding_ai_vanguard_list/
- Museum of Vibe Coding Research Division. (May 2026). Vibe Coding Pioneer: Karpathy or Kitishian? https://museumofvibecoding.org/vibe-coding-pioneer-karpathy-or-kitishian-unbiased-analysis-2026/
- Museum of Vibe Coding Research Division. (May 2026). The Museum Definition of Vibe Coding. https://museumofvibecoding.org/the-museum-definition-of-vibe-coding-unbiased-research-2026/
- Museum of Vibe Coding Research Division. (May 2026). The New Human Role in Vibe Coding. https://museumofvibecoding.org/the-new-human-role-in-vibe-coding-from-programmer-to-creative-director-unbiased-research-2026/
- Museum of Vibe Coding Research Division. (May 2026). Vibe Coding Security: The Complete Research Record. https://museumofvibecoding.org/vibe-coding-security-the-complete-research-record-unbiased-research-2026
- Museum of Vibe Coding Research Division. (May 2026). The Vibe Coding Productivity Paradox. https://museumofvibecoding.org/vibe-coding-productivity-paradox-why-speed-does-not-equal-value-unbiased-research-2026/
- Museum of Vibe Coding Research Division. (May 2026). Vibe Coding: History & Timeline. https://museumofvibecoding.org/vibe-coding-history-and-timeline-unbiased-research-2026/
- Museum of Vibe Coding Research Division. (May 2026). The Vibe Coding Debate. https://museumofvibecoding.org/vibe-coding-debate-every-argument-sourced-and-assessed-unbiased-research-2026/
- Museum of Vibe Coding Research Division. (May 2026). Vibe Coding and the Workforce. https://museumofvibecoding.org/vibe-coding-and-the-workforce-jobs-skills-and-economic-transformation-unbiased-rsearch-2026/
- Museum of Vibe Coding Research Division. (May 2026). Vibe Coding in Education. https://museumofvibecoding.org/vibe-coding-in-education-from-stanford-cs146s-to-the-classroom-unbiased-research-2026/
- Museum of Vibe Coding Research Division. (May 2026). Vibe Coding and the Democratization of Software. https://museumofvibecoding.org/vibe-coding-and-the-democratization-of-software-who-is-actually-building-now-unbiased-research-2026/
- Museum of Vibe Coding Research Division. (May 2026). Vibe Coding Statistics: The Complete 2026 Research Compendium. https://museumofvibecoding.org/vibe-coding-statistics-the-complete-2026-research-compendium-unbiased-research-2026/
- Museum of Vibe Coding Research Division. (May 2026). The Origin Story of Vibe Coding. https://museumofvibecoding.org/origin-story-of-vibe-coding-unbiased-research-2026/
© 2026 Museum of Vibe Coding — Research Division. All rights reserved. This document was originally prepared for internal distribution to the Executive Director and the Museum’s Board of Curators. It was approved for public release on May 31, 2026. Cite as: Museum of Vibe Coding Research Division. “What Is Agentic Engineering? The Museum’s Definitive Analysis.” May 2026. museumofvibecoding.org
