From deterministic IT to living systems that learn, make decisions, and scale
Executive Summary
The adoption of GenAI and agentic architectures does not represent an incremental evolution of IT. It represents a structural break from the traditional operating model. Based on Sharksia’s experience with real-world projects, we have found that the impact is unmistakable: time-to-market accelerates, the cost of change drops structurally, and a new organizational capability emerges: systems that orchestrate decisions, accumulate experience, and evolve with the business; in short, digital frameworks that are progressively capable of self-managing and self-demanding other technological solutions, which they command and can orchestrate.
The central problem of today’s IT is no longer the ability to develop software, but its limited capacity to change with speed and control. The coupling between business, code, and operations turns every change into technical debt, every exception into complexity, and every integration into systemic fragility.
GenAI, combined with declarative programming and agentic architectures, allows us to break this pattern. IT stops executing rigid workflows and begins to orchestrate decisions, acting closer to the event, with semantic traceability and persistent memory. The result is an IT that ceases to be reactive and becomes a designer of intelligent ecosystems that integrate into businesses.
This transformation is strategic, not technological. Although the unit cost of intelligence is decreasing, the volume of decisions, events, and automated processes will grow exponentially. The challenge for companies and their leaders is no longer adopting GenAI, but rather how to govern an IT system that is more autonomous, more powerful, and more deeply integrated into the business.
At Sharksia, we take a clear stance: the future of IT is not programmed solely with code. It is designed as a living system.
1. The Limits of the Traditional IT Model
For decades, IT evolved in layers: better tools, best practices, better infrastructure. In tandem with this, the technology services industry has advanced through incremental improvements: new languages, more productive frameworks, more powerful infrastructure, and more agile methodologies.
However, despite these advances, the mental model of IT has remained virtually intact: deterministic systems, highly coupled, costly to change, and designed to execute predefined logic. And a deterministic process: the business defines requirements; IT implements; operations execute. That model worked as long as change was predictable and variability was limited. But today… change is constant.
At Sharksia, we have observed that the core problem with this model is not “development speed” in and of itself, but rather the interdependence between business, code, and operations: every change becomes technical debt, every exception adds complexity, and every integration introduces fragility.
For all these reasons, at Sharksia we firmly believe that this model no longer scales.
The emergence of Generative Artificial Intelligence (GenAI), combined with declarative approaches and agent-based architectures, does not represent yet another optimization within the same paradigm. It represents a structural shift in the way IT is conceived, built, and operated.
This white paper is neither an academic exercise nor a theoretical projection. It is the expression of a vision built from practical experience and ideas conceived and implemented prior to the massive explosion of these new technologies: experience in real projects, clients who required a connection to context to improve the interpretation of their solutions, experiences with systems operating in real time, and decisions made in contexts where IT must respond with speed, autonomy, and traceability to critical business issues. And subsequently, these new technologies burst fully into the IT world, and we are currently experiencing an explosion where action becomes real, tangible movement.
For all these reasons, for Sharksia, here and now we are not talking about promises. We are talking about a new status quo.
2. GenAI and Agentic AI as a Paradigm Shift
Traditional software is based on a simple premise: all relevant behavior must be anticipated and explicitly coded. This logic has produced robust systems, but they are also rigid, fragile in the face of change, and increasingly costly to evolve.
GenAI enables a profound transition toward AI-native systems, where the focus shifts from procedure to intent, context, and accumulated experience.
In this new model:
Functional change does not necessarily imply technical “refactoring”
Logic is not fully coded; it is orchestrated
The system does not merely execute; it learns
Learning progressively generates experiential collective knowledge, self-management, and self-improvement.
Gartner describes this evolution as the shift toward intent-driven and declarative systems, where value no longer lies in the amount of code produced, but in the ability to design adaptive behaviors. Meanwhile, McKinsey surveys show the adoption and value creation of GenAI in business functions.
For Sharksia, the conclusion is clear: IT is shifting from executing workflows to orchestrating decisions. At Sharksia, we see this as the birth of living systems, capable of evolving alongside the business and, potentially, merging into it as an integral, indivisible part.
3. Declarative Programming + Prompts: Defining the “What”
Prompt-based programming (when properly understood) is not simply “writing text instead of code.” It is a declarative way of expressing intent, rules, and context—of articulating a pure business intent that can be technically recognized. This approach is realized through the integration of tools, functions, policies, and observability.
In an agentic system, the prompt is an engineering component that expresses the business’s “consciousness” (experience) applied to a specific purpose. In this context, the “prompt” constrains behavior, defines criteria, and enables traceable adaptation.
From the Sharksia perspective, the unique value lies not in the ability to write prompts, but in industrializing their design, governance, and evolution. This allows for scaling digital productivity without sacrificing control:
functional changes are managed at the semantic level,
evolution is decoupled from technical refactoring,
quality is sustained through end-to-end traceability,
and the knowledge generated becomes a reusable business asset.
3. Programación declarativa + indicaciones: definir el «qué»
La programación basada en indicaciones (cuando se entiende correctamente) no consiste simplemente en «escribir texto en lugar de código». Se trata de una forma declarativa de expresar intenciones, reglas y contexto, de articular una intención empresarial pura que pueda reconocerse técnicamente. Este enfoque se materializa mediante la integración de herramientas, funciones, políticas y observabilidad.
En un sistema agencial, el prompt es un componente de ingeniería que expresa la «conciencia» (experiencia) de la empresa aplicada a un propósito específico. En este contexto, el «prompt» limita el comportamiento, define criterios y permite una adaptación trazable.
Desde la perspectiva de Sharksia, el valor único no reside en la capacidad de escribir prompts, sino en industrializar su diseño, gobernanza y evolución. Esto permite escalar la productividad digital sin sacrificar el control:
los cambios funcionales se gestionan a nivel semántico,
la evolución se desacopla de la refactorización técnica,
la calidad se mantiene a través de la trazabilidad de extremo a extremo,
y el conocimiento generado se convierte en un activo empresarial reutilizable.
4.2 Maintenance Costs: Cost of Change as a Key Performance Indicator
The real economic turning point lies in maintenance.
In traditional IT, every significant functional change typically involves refactoring, intensive testing, and new deployment cycles. In agent-based architectures, many of these changes are resolved at the semantic level: prompts, rules, policies, and configuration. This reduces the Cost of Change and enables continuous evolution.
At Sharksia, we have successfully executed complete transitions and complex projects—including migrations and rapid adaptations of agent frameworks—within short time windows (50% shorter than the time estimated in traditional IT), while maintaining control and traceability. This is the kind of agility that, in classic software, usually requires rewrites or major refactorings.
McKinsey defines this phenomenon as a structural reduction in the Cost of Change. At Sharksia, we see it as something even deeper: the liberation of IT from its own inertia.
New costs emerge—observability, governance, behavior control—but even considering them, from our point of view, the overall balance is clearly favorable compared to maintaining the model based on classic deterministic software.
4.2 Costes de mantenimiento: el coste del cambio como indicador clave de rendimiento
El verdadero punto de inflexión económico reside en el mantenimiento.
En las TI tradicionales, cualquier cambio funcional significativo suele implicar una refactorización, pruebas exhaustivas y nuevos ciclos de implementación. En las arquitecturas basadas en agentes, muchos de estos cambios se resuelven a nivel semántico: indicaciones, reglas, políticas y configuración. Esto reduce el coste del cambio y permite una evolución continua.
En Sharksia, hemos llevado a cabo con éxito transiciones completas y proyectos complejos —incluidas migraciones y adaptaciones rápidas de marcos de agentes— en plazos muy breves (un 50 % más cortos que el tiempo estimado en la TI tradicional), al tiempo que mantenemos el control y la trazabilidad. Este es el tipo de agilidad que, en el software clásico, suele requerir reescrituras o refactorizaciones importantes.
McKinsey define este fenómeno como una reducción estructural del coste del cambio. En Sharksia, lo vemos como algo aún más profundo: la liberación de la TI de su propia inercia.
Surgen nuevos costes —observabilidad, gobernanza, control del comportamiento—, pero incluso teniéndolos en cuenta, desde nuestro punto de vista, el balance general es claramente favorable en comparación con el mantenimiento del modelo basado en el software determinista clásico.
4.2 Maintenance Costs: The Cost of Change as a Key Performance Indicator
The real economic turning point lies in maintenance.
In traditional IT, any significant functional change typically involves refactoring, extensive testing, and new deployment cycles. In agent-based architectures, many of these changes are resolved at the semantic level: directives, rules, policies, and configuration. This reduces the cost of change and enables continuous evolution.
At Sharksia, we have successfully carried out complete transitions and complex projects—including migrations and rapid adaptations of agent frameworks—within very short timeframes (50% shorter than the time estimated in traditional IT), while maintaining control and traceability. This is the kind of agility that, in classic software, usually requires major rewrites or refactorings.
McKinsey defines this phenomenon as a structural reduction in the cost of change. At Sharksia, we see it as something even deeper: the liberation of IT from its own inertia.
New costs arise—observability, governance, behavior control—but even taking these into account, from our perspective, the overall balance is clearly favorable compared to maintaining the model based on classic deterministic software.
6. Collective Memory: From Data-Centric IT to Experience-Centric IT
Traditional IT, by design, is “amnesic”: it records logs, at best stores data, and executes logic, but rarely transforms these elements into persistent operational learning, as it fails to turn them into “stories” or memory.
Agentic architectures, on the other hand, introduce the distinctive feature of creating, building, and reusing memory, with the capacity to be persistent and based on comprehensive event management.
This memory is considered at different levels:
Episodic memory: what happened and in what context
Semantic memory: emerging patterns and rules
Procedural memory: which decisions work best, to accelerate future decisions.
An event, with its context and end-to-end traceability, generates what we at Sharksia call a “memory thread.” When these memories are versioned, shared, and reused, what we at Sharksia call collective memory emerges, with its clear potential to evolve into collaborative enterprise memory. This memory constitutes true consciousness,This mechanism creates a unique cumulative advantage: the system improves with use. In business terms, it transforms operations into a knowledge asset. This idea aligns with the trend of “software that composes experience” and with the trend of systems shifting from execution to learning.
McKinsey refers to “organizational learning systems.”
For Sharksia, this is a point of no return: the system improves with use. And even to the point of self-management.
7. Semantic traceability and living governance
In agentic systems, traceability ceases to be purely technical and becomes semantic:
What event occurred
Which elements of the environment were relevant.
What reasoning was applied
What decision was made
What result was obtained
This capability transforms the relationship between IT and business. Decisions are understood, audited, and adjusted quickly.
The direct impact is a tangible reduction in time-to-market and a This mechanism creates a unique cumulative advantage: the system improves with use. In business terms, it transforms operations into a knowledge asset. This idea aligns with the trend of “software that composes the experience” and with the trend of systems shifting from execution to learning.
McKinsey refers to “organizational learning systems.”
For Sharksia, this is a point of no return: the system improves with use. And even to the point of self-management.
7. Semantic traceability and living governance
In agentive systems, traceability ceases to be purely technical and becomes semantic:
What event occurred
What elements of the environment were relevant.
What reasoning was applied?
What decision was made?
What result was obtained?
This capability transforms the relationship between IT and the business. Decisions are understood, audited, and adjusted quickly.
The direct impact is a tangible reduction in time to market and a 9. Conclusion — The New IT Contract
The true value of prompt + Agentic AI lies not only in reducing costs or accelerating delivery, but in enabling autonomous systems with memory, traceability, and the ability to continuously learn—systems that transform IT from a reactive executor into a living, adaptive, and strategic system.
GenAI applied to technology services is neither a fad nor an additional layer on top of existing systems. It is the tipping point that marks the end of one operating model and the beginning of another.
The real decision is no longer technological. It is strategic.
Organizations must decide whether to continue optimizing deterministic systems that are increasingly costly to change, or to move toward intelligent ecosystems capable of learning, adapting, and evolving with control and purpose.
At Sharksia, we are not observing this transformation from a distance. We are helping to build it.
For Sharksia, the future of IT is not programmed solely with code. It is designed as a living system.
Key Insights:
The current state of IT cannot scale: the main bottleneck is no longer development capacity, but the Cost of Change.
GenAI + declarative programming + Agentic AI are redefining the role of IT: from an executor of deterministic workflows to an orchestrator of controlled decisions.
“Hive”-style architecture enables scaling intelligence without increasing technical complexity.
Collective memory turns operations into a strategic asset that improves with use and enables self-learning.
Semantic traceability enables responsible autonomy: decisions that are explainable, auditable, and governable.
The unit cost of intelligence is falling, but the strategic impact of IT is growing: governing consumption and value is a board-level priority.
References cited in the paper
Gartner (June 25, 2025). “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027” (press release).
Reuters (June 25, 2025). “Over 40% of agentic AI projects will be scrapped by 2027, Gartner says.”
McKinsey (June 27, 2023). “Unleashing developer productivity with generative AI.”
McKinsey (May 30, 2024). ‘The state of AI in early 2024: Gen AI adoption spikes and starts to generate value’.
a16z (Nov. 12, 2024). ‘LLMflation: LLM inference cost is going down…’.
Microsoft Learn (Dec. 5, 2025). ‘Governance and security for AI agents across the organization’ (Cloud Adoption Framework).
Microsoft Learn (January 16, 2026). “Copilot Studio security and governance.”
Microsoft Copilot Blog (December 15, 2025). “What’s New in Microsoft Copilot Studio: November 2025” (agent governance & production readiness).
Appendix 1 (optional for publication; currently being refined)
How to Measure Transformation: Sharksia KPIs
A common mistake is to measure GenAI solely based on “time savings” in isolation. The transformation driven by GenAI should focus on measuring the ability to change IT in a controlled manner.
At Sharksia, we propose KPIs across six dimensions:
10.1 Time-to-Market
Functional change lead time (request → production).
% of changes delivered without major technical refactoring.
Onboarding time for a new use case (first value).
10.2 Cost of Change
Cost of Change Index (CCI) = total cost of changes / number of functional changes.
% of changes made via prompt/policy vs. code changes.
Post-change defects per 100 changes (evolution quality).
10.3 Autonomy and Orchestration
Event-to-action latency (time).
% of operational decisions executed by agents with defined supervision.
Human-in-the-loop intervention rate per workflow.
10.4 Memory and Learning
% of decisions that explicitly reuse memory (episodic/semantic).
Average exception resolution time (functional MTTR).
Incident recurrence rate due to the same cause (should decrease).
10.5 Governance and Traceability
% of decisions with full semantic traceability.
Average audit/forensic time per incident (should decrease).
Security incidents due to excessive permissions (should decrease).
10.6 Strategic Impact
% of IT initiatives linked to business metrics (revenue, cost, experience).
Business–IT satisfaction (internal NPS).