By Jacques Bughin
The rapid rise of large language models and generative AI is redefining how software is built and scaled. Jacques Bughin explores how vibe and agentic coding shift value from writing code to orchestrating intelligent systems, signalling a transformation that extends beyond software into broader business strategy and competitive advantage.
The emergence of large language models (LLM) and generative AI programming tools has boosted a paradigmatic change in software development.
Today, a large portion of programming is done through AI assistant, with claims of significant productivity and quality gains. The spread of adoption is unusually fast, with share of code generation doubling every year, driven also by new practices such as vibe coding, centering around a prompt-driven approach to coding, to rapidly turn ideas into working prototypes or features.
The consequence of this paradigm shift, is not only that the software engineer will soon be replaced by a vibe-driven user. It also means that software barriers to entry will plummet and commoditize. Competitive advantage will have to come outside of code; from somewhere else such as trust, distribution, domain expertise and exécution. Software also become consumer goods as they can be adapted on demand, and possibly automaticially through agentic coding.
Software also become consumer goods as they can be adapted on demand, and possibly automaticially through agentic coding.
Finally, even for companies and people with limited exposure to software, the consequence is to reframes complexity as a temporarily unsolved problem, awaiting its own “AI agent and vibes”. Domains like law, medicine, logistics, or financial engineering—all with long-standing barriers of expertise—may equally face disruption from tools that automate away their trickiest challenges. In this new world, value moves from knowing “how” (implementation) to knowing “what” and “why” (problem selection), But is this new world as close as it stands? Here is what to have in mind-from now on.
The democrazation of software development
Historically, software development has been the domain of technically trained professionals, constrained by steep learning curves and the need for mastery in programming languages and environments. Early innovations such as graphical integrated development environments (IDEs) eased usability compared to command-line interfaces, but they retained the core requirement: users still had to code.
No-code and low-code
To address the exclusionary nature of traditional software engineering, successive waves of “democratization” emerged. No-code and low-code platforms introduced visual, drag-and-drop interfaces with pre-built logic components, aiming to empower non-programmers. However, despite enabling broader participation, these tools were limited in flexibility, customization, and scalability. This “visual programming” era still required users to work within rigid templates and faced challenges when applications grew in complexity or required deeper logic control.
Recent breakthroughs in Generative AI, particularly Large Language Models (LLMs) trained on code (e.g., Codex, CodeGen), have radically changed this landscape. These models can now generate functional software components from natural language descriptions—blurring the boundary between technical and non-technical contributors. This capability marks a significant leap from procedural or visual programming to intention-driven development.
Vibe coding
The new emerging concept of “vibe coding”, popularized in early 2025 by AI researcher Andrej Karpathy, where humans guide powerful AI models—mostly using natural language—to generate, debug, and refine code, exemplifies this shift: instead of writing lines of code, users express their desired outcome in natural language, and AI systems translate these “vibes” into functional applications.
The recent hype comes from the explosion in AI tooling. recent advances in LLM-powered coding agents—such as Cursor Composer, JetBrains, and GitHub Copilot—have made it possible to create entire applications by talking to the AI, rather than manually coding every feature. Also, startup and industry are qucly buy into this new tend. Prominent tech industry figures and accelerators like Y Combinator have publicly declared vibe coding “the new dominant way to code,” with 25% of startups reporting more than 95% of their codebase generated by AI.
Platforms like Replit saw astonishing growth (from $10M to $100M ARR in six months), signaling massive adoption and excitement in both consumer and business spaces. Around 75% of Replit users never write a single line of code. The CEO of Robinhood recently declared that nearly 100% of their developers use AI code editors, and around 50% of new production code originates from AI.
On the enterprise side, AWS has entered the ring with Kiro, an agentic IDE in preview that breaks natural-language prompts into structured blueprints, enforces steering policies, and auto-generates verification tests—effectively embedding production-grade rigour directly into vibe-driven flows. Meanwhile, Replit, Visa, Vanguard, and Choice Hotels are already piloting vibe-centric systems, reporting 40% faster UI development cycles and enhanced collaboration between engineers and non-technical stakeholders.
Vibe coding maturity cycle
The messiness of vibe coding
The market for vibe/ AI-Codegenerators was evaluated to USD 4.5 billion in 2023 to become ten times larger by 2030. In aggregate, companies, form Windsurf, to lovable and gityHUb are already worth 50 USD bilion + . But, while vibe coding promises broader participation in software creation, adoption is mixed by maturity. Some companies treat vibe coding as an innovation engine; others warn that hallucinations and package-misnaming errors—coined “slopsquatting”—pose severe risks
In fact, as seen in the early deployements of LLMs, – LLM-generated code can inherit biases, security vulnerabilities, and inefficiencies from the training data. Users often lack visibility into how code is generated, making transparency and explainability key challenges—especially in high-stakes or regulated environments.
There is also a looming risk of vendor dependency and AI system lock-in, as organizations integrate proprietary generative AI tools into their workflows. This necessitates robust governance frameworks, including technical review protocols, domain-specific risk classifications, and phased adoption strategies. For instance, internal applications with low business risk can serve as testbeds before expanding to customer-facing or mission-critical systems.
Christensen theory of disruption
But, disruption rarely begins with elegance. It starts with functional access, then layers reliability. Vibe coding mirrors this: ugly at the start, but full of potential foir those seeing the upgrade path from inferiortiy to disruption, as elegantly discovered by Clay Christensen.
In the recent past, GUIs were seen as slow, cluttered alternatives to the command line — until they unlocked mass adoption. APIs began as fragile connectors, only to become the scaffolding of platforms and microservices. Cloud computing started as an ops headache — then ushered in a DevOps revolution.Promoters like Microsoft, Apple, Amazon, and Google were critical to scaling these interfaces beyond their early mess. Similarly, today’s big cloud players and AI-first startups are the primary promoters of vibe and agentic coding. Their infrastructure, standard-setting, and distribution power will be decisive in whether the chaos resolves into dominance.
The five conditions for vibe coding to evolve in disruptions are slowly being met
- Determinism and auditability: Prompt logs, version control, and reproducibility must mature. Without these, vibe remains chaos. Leading platforms such as GitHub Copilot and Cursor now incorporate prompt logs, code history features, and are rolling out reproducibility safeguards
- Agentic refactoring & testing: Tools like Qodo and Devin must become reliable backstops that ensure code correctness. Reports highlight a “burgeoning ecosystem” of agentic tools (e.g., Replit’s Codeium, agentic capabilities in Cursor) acting as automated reviewers and testers. While reliability is not perfect, dependency on these agents for maintaining codebase correctness is increasing rapidly, especially among tech-first startups and digital agencies
- Agent orchestration IDEs: IDEs like Cursor and Kiro must become full orchestration hubs — blending prompt history, test logs, and repo memory. e IDE landscape is shifting: developers are flocking to environments (Cursor, Replit, Kiro) that integrate prompt histories, live testing, repo memory, and agent “marketplaces.” Cursor, for example, saw its user base grow to over 1 million—with one-third paying—largely due to its orchestration features.
- Traceability & governance: Integration with CI/CD pipelines and audit systems is essential for enterprise scale. Uptake is strongest in startups and digital natives, but more conservative sectors (health, finance) are already piloting integrations with CI/CD and audit pipelines
- Swarm agents: Specialist bots must handle packaging, performance, security, and compliance automatically.
These conditions also echo prior disruption conditions from GUI and cloud: a layer of abstraction must normalize chaos, professionalize interfaces, and absorb complexity. Cloud moved to a structuration of I/P/SAAS creating standardized infrastructure and service layers for deploying, managing, and consuming software. GUI moved to a structuration of standardized interaction paradigms and design metaphors
Strategic lens
The advent of vibe coding and agentic AI represents more than a technical advance—it signals a foundational shift in how software is created, maintained, and scaled.
The advent of vibe coding and agentic AI represents more than a technical advance—it signals a foundational shift in how software is created, maintained, and scaled. Much like previous technology revolutions, it begins with awkward prototypes, imperfect outputs, and early adopters experimenting at the fringes. But as Clayton Christensen taught, disruption never starts with elegance. It starts where incumbents are least likely to pay attention, then climbs the value chain as speed, reliability, and reach improve. What begins today as an experimental tool for developers may soon become the operating system of modern business.
CEOs should view this moment through that exact lens. The disruption is already underway. Code generation by AI is doubling annually. Major tech firms like GitHub, Amazon, and Replit are accelerating orchestration tooling that allows teams to go from idea to deployment through prompts and agent collaboration—cutting development time by up to 40%. The entry point may appear technical, but its impact is strategic: it redefines the source of competitive advantage. Software is no longer a scarce asset. It is becoming abundant, cheap to produce, and increasingly commoditized. In this new environment, value shifts from owning the code to orchestrating how it’s created, validated, and governed. As such, your organization must transition from being a software builder to either a producer of orchestration systems or a sophisticated consumer of AI-produced software.
If your firm is in the business of software, this is the time to reposition. AI is collapsing the cost and time required to build features, apps, and services. Traditional R&D pipelines will become less defensible unless they’re coupled with governance architectures, agent supervision, and trust layers that make AI output reliable, auditable, and tailored to critical domains. The product is no longer the code—it is the environment that allows teams to safely and rapidly produce code that works. Investment must shift toward agentic orchestration, intelligent testing, and cross-agent collaboration environments. You’re not just competing on features anymore—you’re competing on how fast, how safely, and how intelligently your agents can adapt code to user needs.
If your firm is not in the software business, this shift is no less consequential. Historically, creating software required skilled developers, long cycles, and large budgets. But with vibe coding, natural language prompts can drive application development. The bottleneck of coding disappears, replaced by a new frontier: selecting the right problems, framing them effectively, and validating outcomes. In this world, your organization can become a software producer without traditional engineering teams—or remain a consumer of standardized tools, at risk of being locked out of differentiated capabilities. The choice is strategic. To lead, firms must cultivate internal orchestration capacity—not to write code per se, but to shape how AI agents do. Teams in product, legal, finance, or logistics may become creators by guiding agents. But without a coherent governance model and cross-functional prompt fluency, this capability will remain fragmented and underutilized.
Beyond the technical and strategic shifts lies a deeper organizational question: who in your company needs to be empowered to build, orchestrate, and validate digital systems? As AI takes over the “how,” success will depend on those who best define the “what” and the “why.” Companies must cultivate prompt designers, orchestration leads, and agent supervisors across all business units. This change affects training, recruitment, incentives, and culture. It also alters who holds influence: the ability to ask the right question and steer agentic systems will matter more than knowledge of syntax or frameworks. If previously only software engineers built software, now every domain expert can become a producer—provided they are equipped with the right interfaces and governance layers.
This transformation will not happen all at once. But the signals are clear. Firms like Robinhood report nearly 100% of developers using AI editors, and over half of new production code being AI-generated. Replit’s user base is now majority non-coder. Cursor and Kiro are embedding orchestration, traceability, and agent marketplaces directly into development environments. The tooling is catching up. The maturity conditions Christensen warned about—reliability, auditability, governance, ecosystem—are being met faster than most expect.
What matters now is whether your organization learns to orchestrate AI agents effectively, govern their outputs safely, and redeploy software capabilities in every domain—not just in IT.
As CEO, your role is not to manage code. It is to recognize that code has ceased to be the strategic asset it once was. What matters now is whether your organization learns to orchestrate AI agents effectively, govern their outputs safely, and redeploy software capabilities in every domain—not just in IT. This is not the time to wait. It is the time to learn, pilot, and reposition. Because soon, software won’t be something your company builds or buys. It will be something it behaves through—constantly adapted, agentically maintained, and invisibly embedded in how you operate.
The question is no longer whether vibe and agentic coding will arrive. It is: when they do, will your company be leading the orchestration—or watching from the edge, unable to control what it consumes. The AI is full of surprises, but one thing is sure – disruption is about to start.
Jacques Bughin
