Essential
Required for the site to work. Always on.
Chief Strategy Officer · Thinkia
David Alejano
David Alejano is Chief Strategy Officer at Thinkia, the AI-native consultancy. He works where strategy meets architecture: deciding what an organisation should build, in what order, and what it should govern before it builds anything at all.
He writes the firm's public thinking. The articles published under his byline on Thinkia's Thoughts, and the whitepapers below, deal with the same recurring problem from different angles — the distance between adopting AI and being AI-native. Governance as design rather than paperwork. Agentic orchestration that a board can sign off on. Legacy systems that get modernised instead of rewritten. Shadow AI governed instead of banned.
He speaks on it too. At TheFringe/LABS in Madrid he put the argument in one line: the challenge is no longer access to technology, it is fragmentation.
Strategy is the bridge between where we are and where we need to be. I craft the roadmap that transforms ambitious visions into actionable plans, ensuring every decision aligns with our long-term goals while navigating the complexities of digital transformation.
01
What I work on
Five lines run through everything published here. They are the categories of the Thoughts archive, not a positioning exercise.
- Innovation Insights 54
Where the models are going, and which of it survives contact with an enterprise.
- Platforms & Engineering 53
Gateways, agent orchestration, cost control and the architecture underneath.
- Automation 17
What a system can be trusted to decide alone, and how you keep it governable.
- Data & Analytics 10
Turning scattered data into something a system can reason over, not merely search.
- Product & Experience 10
The interface is where governance either shows up or quietly disappears.
02
Whitepapers
Long-form arguments, each one a document you can take to a committee. Free, no form.
-
AI-Native vs AI-Adopted. Adopting AI is not being AI.
This whitepaper argues that adopting AI is not the same as being AI, and that the difference sits in ontology, talent and speed rather than in the budget. It defines the two archetypes and the question that organizes each one, explains the path dependency that produces adopters by inertia, details the five structural differences across ontology, talent, methodology, stack and speed, examines the four organizational brakes that keep an adopter from becoming native, offers five signals any transformation buyer can read in one meeting, and states honestly where the adopted model is sufficient.
-
The company with memory. Storing is not remembering.
This whitepaper argues that companies do not have operational memory: they have amnesia governed by folders, and a document repository is not memory. It covers corporate amnesia and its symptoms, why expanding the repository worsens the problem, the four properties of operational memory (accessible in the workflow, contextual, traceable, alive), the three dimensions of recall (episodic, semantic, procedural), the five-layer architecture of a company that remembers, the key-expert risk measured by the bus factor, and three anonymized cases that show the order of magnitude of the return.
-
The company as an intelligent system. From archipelago to organism.
This whitepaper describes the shift from a company that has AI to a company that is an intelligent system. It covers the archipelago problem, systems that calculate well separately and think poorly together; why integrating applications moves data but not judgment; the five properties of the enterprise organism, from central nervous system to governance; the architecture in four layers of intelligence; the five-phase construction path, from honest diagnosis to industrialized learning; the four pathologies on the risk map; and the four KPIs that tell whether the company thinks.
-
Shadow AI. Govern it, don't ban it.
This whitepaper argues that the invisible risk is not that employees use AI, but that they use it where the company cannot see it. It covers what Shadow AI is and why it is already inside, the anatomy of the phenomenon and the three forces that feed it, the inventory of real risks from data leakage to unaudited decisions, why a blanket ban keeps the risk and removes observability, the three-move governance framework (make visible, govern, enable), the CISO paradox of turning security from wall into channel, and the four indicators that show whether the framework works.
-
Legacy isn't rewritten. Brownfield AI.
This whitepaper argues that legacy systems are not the problem — the classic approach of rewriting them is. It covers why the classic approaches fail, how to use the spec as the key migration asset, the anatomy of AI-enabled reverse engineering, a five-phase brownfield migration method, two real implementation cases, a risk map for delivery, and the success KPIs that prove value.
-
AI Dashboards. Ask, don't filter.
This whitepaper argues that the new business intelligence is a conversation with data, not a dashboard you interrogate. It covers why nobody opens the dashboard, the anatomy of traditional BI, how to invert the flow, the three AI-native layers—natural conversation, proactive anomaly, generated narrative—a conversational architecture, cases by function, what breaks the demos in production, and the adoption KPIs that matter.
-
AI Ethics & Governance. Beyond compliance.
This whitepaper shows why the EU AI Act is more than a compliance checklist: how AI has left the lab, what operational governance requires, the Thinkia AI Compass Framework, applied ethics in practice, and a staged path to realize governance while unlocking competitive advantage.
-
AI–SDLC. From code to intent.
This playbook describes how to industrialise AI inside your SDLC: the five delivery phases, the three master specification files, green/amber/red risk tiers, and how to run a Sprint 0 plus Stage 1 MVP without unmanaged copilot sprawl.
03
Articles
The most recent, under my byline. The full archive is open, with no form in the way.
- Prompt Engineering Limits: Why More Context Isn't Always Better for AI A recent study reveals the hard limits of prompt engineering, showing more context doesn't always improve AI model accuracy. This underscores the need for empirical testing and AI performance validation over relying on generalized claims. For specialized tasks, fine-tuning may be essential.
- Hierarchical AI Systems: The Enterprise Playbook for Cost-Efficient LLMs Hierarchical AI systems are the key to unlocking cost-efficient LLM inference at scale. By dynamically routing tasks between small edge models and large cloud models, enterprises can slash costs and reduce latency. This playbook covers the strategic shift towards adaptive model inference and how to prepare your architecture for this new reality.
- LLM Inference Optimization: Beyond Hardware with Vector Indexes New research shows LLM inference optimization can be achieved algorithmically, reducing AI inference cost by up to 82% without new hardware. This makes edge AI and wider model deployment more viable. We explore the strategic implications for enterprise leaders.
- The Future of AI Inference Hardware: Beyond the GPU Monopoly Google's integration of TPUs with vLLM signals a major shift in the AI inference hardware landscape, challenging Nvidia's dominance. We analyze the strategic implications for enterprise AI infrastructure and why CIOs must prepare for a multi-vendor, performance-driven future for workloads like RAG.
- Human-in-the-Loop AI: A Practical Blueprint for Regulated Industries Discover the human-in-the-loop AI cascade model, a pragmatic approach for enterprise AI in regulated industries. This blueprint for document classification and other high-stakes tasks balances automation with expert oversight, enhancing AI governance and reducing risk.
- Secure On-Premises AI: Why the Mainframe Is Making a Strategic Comeback Secure on-premises AI is gaining traction in regulated industries. IBM's latest RAG architecture on LinuxONE shows how enterprises can leverage AI without compromising enterprise AI security or data sovereignty. We analyze this emerging trend and its implications for CIOs.
04
Talks
Where the argument has been made in person.
-
TheFringe/LABS
TheFringe/LABS brought together executives and investors at the AEDAS Homes flagship to debate AI governance, agentic orchestration, and investment. Thinkia shared how the Synapse Agentic Platform turns the democratisation of AI into real business value under a solid governance framework.
Get in touch
For a conversation about AI strategy, governance or where to start, the contact form reaches me and my team. For everything else, LinkedIn.
Articles, whitepapers and talks on this page are read from the site's own content at every deploy, so the list is never out of date.