Vibe coding allows for previously unimaginable capabilities to be built and orchestrated in near real time. An employee spots a need, describes it in natural language, and has a working prototype within hours. A team could respond to a competitor move or a regulatory change with functioning software super quickly. How that software gets evaluated, deployed and governed is key.
Now of course it’s not just humans doing the building. Agents, whilst having hotly debated abilities to truly reason or handle nuanced human processes, allow for algorithmic, semi-autonomous capabilities that can respond to ecosystem dynamics: upstream suppliers, downstream customers, regulators, competitors. The asymmetries this creates, as I discussed in my sleepwalking piece last year, are profound. Organisations that harness this well gain compounding advantages. Those that don’t accumulate compounding pain.
The Cambrian Explosion of AI generated point solutions
If vibe coding and agentic capabilities are adopted tactically, by individual employees or agents acting in silos, you don’t get innovation. You get a Cambrian-style explosion of point solutions that don’t talk to each other, duplicate effort, and add cost and complexity at an extraordinary rate. This could be like spreadsheet hell all over again, but with far worse consequences! (Rogue formulas were one thing but agentic AI is far more powerful).
GitClear’s analysis of 211 million lines of code shows refactoring collapsing from 25% of changed lines in 2021 to under 10% by 2024, as developers accept AI output without the iterative cleaning they would apply to human-written code. Open-source maintainers are shutting their doors to external contributions because the signal-to-noise ratio from AI-generated submissions has become unworkable.
The industry conversation has caught up to part of this. WaveMaker predicts a 2026 reset from vibe coding experiments to architecture-first, governed tools. Gartner projects that 40% of enterprise applications will embed AI agents by the end of 2026, up from less than 5% in 2025.
The governance issue is now critical, but we’d argue it’s happening at the wrong level.
The Missing Layer: Ecosystem, Not Just Enterprise
Almost every article, analyst prediction, and vendor pitch frames this as an internal enterprise governance problem. How do we govern what our developers build? How do we ensure code quality? How do we manage our agent fleet?
These are necessary questions, but they’re insufficient. They treat the organisation as a closed system. It isn’t.
When your agents interact with supplier APIs, when your vibe-coded integrations touch customer data, when your AI-generated code calls open-source libraries whose maintainers have just shut down contributions, when your architecture decisions have environmental consequences that your sustainability reporting doesn’t capture: you’re operating in an ecosystem, not an enterprise.
This is the insight that Ecosystem Architecture, which I first wrote about nearly 10 years ago, was built around. Your internal capabilities don’t exist in isolation. They’re nested within supply chains, regulatory environments, community ecosystems, and natural systems. The decisions you make ripple outward, and the decisions others make ripple inward. Pretending otherwise is how you end up with architectures that look rational internally but are fragile, extractive, or unsustainable at the system level.
The Cone of Possibilities
As I’ve discussed previously, the GenAI and agentic era doesn’t point in a single direction. It unlocks a cone of possibilities.
At one end: massive cost and externalities, enshittification of products and services, deeper monopolistic conditions, and environmental damage that makes the current AI infrastructure buildout look modest.
At the other: revolutionary, efficient, hyper-personalised capabilities that genuinely serve organisations and individuals, built on architectures that are conscious of their broader impacts.
Most organisations will land somewhere between these extremes. Where they land depends on whether they navigate strategically or follow the herd.
Following the herd currently means adopting whatever the hyperscalers are selling, embedding AI into everything without asking whether it should be there, and treating governance as a compliance afterthought. It means vibe coding without architecture, agents without guardrails, and innovation without awareness of downstream consequences.
Navigating strategically means something different entirely.
What is Dynamic Architecture?
Dynamic Architecture is the governed, intentional response to a world where technology capabilities can be created, modified, and orchestrated in near real time.
It builds on Ecosystem Architecture and the Pragmatic Ecosystems methodology. It recognises that in a world of vibe coding and agentic proliferation, architecture can no longer be a static set of blueprints reviewed quarterly. It needs to be a living capability: responsive, adaptive, and bounded by clear principles and guardrails.
In practical terms, Dynamic Architecture means:
Guardrails that enable, not gates that block. Traditional architecture governance assumes change is slow and can be reviewed before deployment. That assumption is dead. Architecture principles, patterns, and constraints need to be embedded into the tools and environments where creation happens, not applied after the fact. Think of it as the difference between a speed limit sign (which people ignore) and a road that is physically designed for safer speeds.
Ecosystem awareness as a first-class concern. Every dynamically created capability has upstream dependencies and downstream effects. Who does it rely on? Whose data does it touch? What are the environmental costs? What happens when the open-source library it depends on stops being maintained? These aren’t edge cases. They’re architectural fundamentals.
Viability in the cybernetic sense. Drawing on Stafford Beer’s Viable System Model, Dynamic Architecture asks: does your organisation have the feedback loops to detect what’s being created, understand its impacts, and respond when things go wrong? Given an individual employee can now vibe-code a new integration in an afternoon, the answer for most organisations is no!
Sustainability as an architecture decision. Every piece of dynamically generated code consumes resources: compute, energy, water, materials. Every agent interaction has a cost. A paper in Scientific Reports on the environmental comparison of AI and human programming found that model selection is critical to minimising environmental impact, and that the productivity gains from vibe coding could significantly increase the total volume of software produced, amplifying the environmental footprint. Dynamic Architecture treats these impacts as architecture concerns, not externalities to be reported on later.
What This Means in Practice
The alternative to Dynamic Architecture is what we’re seeing unfold in real time: a second wave of technical debt, larger and faster-accumulating than anything the industry has seen before. The first wave came from decades of tactical enterprise IT decisions. This wave will come from months of tactical AI adoption.
Organisations that get this right will:
- Create more value from AI and agentic capabilities, because those capabilities will be coherent, not contradictory
- Avoid the re-architecture costs that follow from letting hundreds of disconnected vibe-coded solutions accumulate (costs that early evidence suggests exceed the initial development savings)
- Maintain genuine optionality, because well-governed dynamic capabilities can be reconfigured, whereas a tangle of point solutions locks you in
- Understand and manage their broader impacts: on data, on people, on the environment, on the ecosystem they depend on
Organisations that don’t will find themselves in a familiar but accelerated trap: moving fast, accumulating mess, and spending the next decade cleaning it up. Only this time, the mess will be larger, more distributed, and harder to see.
Building on Foundations
Dynamic Architecture isn’t a new framework invented to chase a trend. It’s the natural evolution of thinking we’ve been developing for nearly a decade.
Ecosystem Architecture (2017) established that technology architecture doesn’t stop at the enterprise boundary: organisations participate in ecosystems rather than sitting at their centre. The Pragmatic Ecosystems methodology connected business and technology architecture to real-world impacts using cybernetics, Doughnut Economics, and Circular Economy thinking. On the measurement side, my work with the Green Web Foundation on DIST, the Digital Impacts Schema and Taxonomy, is about seeing technology’s share of those impacts beyond carbon.
Dynamic Architecture is what happens when you apply these foundations to a world where technology creation is no longer constrained by the speed of human typing.
The Conscientious Question
At the heart of this is a question that the technology industry consistently fails to ask: how do we collectively build systems that improve our world, rather than just further lining the pockets of the tech industry and creating yet more monopolistic conditions?
Vibe coding and agentic AI are tools. They can be directed toward genuinely useful, efficient, and equitable outcomes. They can also be directed toward more extraction, more waste, more complexity, and more concentration of power. The direction depends on architecture: on the deliberate, governed, ecosystem-aware decisions about what gets built, how, and at what cost.
Dynamic Architecture, grounded in Pragmatic Ecosystems, is the responsible response to where we find ourselves today. Not anti-innovation. Not anti-AI. But fiercely pro-intentionality, pro-governance, and pro-awareness of the systems we’re all embedded in.
The organisations that understand this won’t just navigate the next few years more successfully. They’ll create agile capabilities that are truly differentiating and allow them to remain relevant in the 2030s.
Oliver Cronk is a fractional CTO and chief architect and the founder of Cronk Advisory. He coined Ecosystem Architecture in 2017, founded the Architect Tomorrow community in 2019, and is working on the Pragmatic Ecosystems approach.
This is the first in a series exploring Dynamic Architecture.
AI use: AI augmented this piece rather than replacing the author. Claude helped with research synthesis and drafting; the ideas, framing, judgement and any errors are mine.