# Donovan Crewe > Writing by Donovan Crewe, CTO at Lumenalta, on AI-native engineering, Engineering Intelligence, agentic software delivery, AI governance and evaluation. Donovan Crewe is Chief Technology Officer at Lumenalta. Every article is available as Markdown at /article/.md; the full corpus is at https://dcrewe.com/llms-full.txt. ## Articles - [The Harness Isn’t Dead](https://dcrewe.com/article/the-harness-isnt-dead.md): The hand-holding is - [The Two-Year Migration Plan Is Now the Risk](https://dcrewe.com/article/the-two-year-migration-plan-is-now-the-risk.md): Why agentic delivery changes the economics of data platform migration - [The Meat Proxy Paradigm](https://dcrewe.com/article/the-meat-proxy-paradigm.md): AI coding agents let engineers produce sophisticated software they may not fully understand. This article explores the 'meat proxy' problem and the growing gap between execution capability and decision capability, arguing that AI-native engineers should be measured by whether they become better at engineering, not by how much code they delegate. It introduces meaningful AI usage and Engineering Intelligence as the way organisations turn every AI interaction into learning. - [When the Engineers Building AI Start Leaving](https://dcrewe.com/article/when-the-engineers-building-ai-start-leaving.md): As safety researchers leave frontier AI labs for independent evaluation organisations, this article looks past the sensational headlines to the real engineering problem: our ability to understand and control increasingly autonomous systems is not keeping pace with their capabilities. It argues that governability must become a first-class architectural requirement, built through scoped credentials, isolated environments, observable actions and reliable ways to stop execution. - [Why Every CTO Needs an Intelligence Strategy](https://dcrewe.com/article/why-every-cto-needs-an-intelligence-strategy.md): Most organisations have an AI strategy focused on models, tools and governance. This article argues that CTOs need an Intelligence Strategy instead: a deliberate plan for how intelligence enters the organisation, how it circulates, how it compounds and how you know it is improving. It reframes the CTO's role from selecting technology to cultivating Engineering Intelligence. - [Engineering Intelligence](https://dcrewe.com/article/engineering-intelligence_-the-new-strategic-asset.md): As code, cloud infrastructure and frontier models become commodities, this article argues that the lasting strategic asset of an engineering organisation is Engineering Intelligence: the capability that emerges when engineers, AI and organisational knowledge operate as a continuously learning system. It connects organisational memory, personalised coaching, evaluation and agentic delivery as investments in the same asset. - [It’s Time to Admit Software Engineering Has Fundamentally Changed](https://dcrewe.com/article/its-time-to-admit-software-engineering-has-fundamentally-changed.md): AI makes it far harder to know whether a change to a prompt, model, agent or workflow actually made things better. Using the Cat Council as an example of independent judges whose disagreements are reconciled into a shared definition of quality, this article makes the case for Evaluation Engineering: continuously measuring whether humans, AI and the systems around them are genuinely improving. - [AI Adoption Is Not the Problem](https://dcrewe.com/article/ai-adoption-is-not-the-problem.md): Organisations have invested heavily in rolling out AI tools but very little in helping engineers get better at using them. This article introduces AI Capability Engineering, a discipline for observing how engineers work with AI, surfacing both friction and excellence, and delivering personalised coaching at scale so the organisation learns collectively rather than one engineer at a time. - [Why AI Isn't Scaling in the Enterprise](https://dcrewe.com/article/why-ai-isnt-scaling-in-the-enterprise.md): Impressive individual AI demonstrations rarely translate into enterprise-wide transformation, and the reason is cost. This article explains how token budgets, restricted models and truncated context quietly undermine AI effectiveness, and argues for Intelligence Efficiency: using persistent project state and organisational memory so AI never has to reason about the same thing twice. - [Why 2026 Is The Year Software Delivery Stopped Being Human-Centric](https://dcrewe.com/article/why-2026-is-the-year-software-delivery-stopped-being-human-centric.md): The SDLC was designed around humans collaborating through sequential phases, and around engineers learning by doing the very work AI is now removing. This article argues that AI turns delivery from sequential to concurrent, makes engineering judgement the scarce resource, and introduces the AGE lifecycle: agents execute, governance protects, engineers elevate. - [The AGE Engineering Principles](https://dcrewe.com/article/the-age-engineering-principles.md): The AGE Lifecycle exists because software delivery is increasingly driven by autonomous systems operating under human governance. This article sets out nine principles for AI-native engineers, from intent over implementation and constraints that create scale to continuous evaluation, context as infrastructure and accountability that remains human. - [The Dim Factory](https://dcrewe.com/article/the-dim-factory.md): Between treating AI as a developer convenience and chasing fully autonomous software dark factories sits a more practical model. This article describes the dim factory: governed pipelines where agents handle high-volume work within enforced boundaries, humans design the system and oversee critical decisions, and autonomy increases gradually as models become more reliable. - [The Invisible Revolution Reshaping Enterprise Operations](https://dcrewe.com/article/the-invisible-revolution-reshaping-enterprise-operations.md): Beyond the chatbots and copilots, generative AI is becoming the hidden middleware of enterprise workflows across HR, legal, finance and support. This article explores how it is quietly rewiring operations, shifting people from consuming raw material to curating and approving outputs, and why that invisible plumbing will decide who runs faster, leaner and more compliantly. - [The Evolving SDLC](https://dcrewe.com/article/the-evolving-sdlc_-navigating-ais-impact-on-software-delivery.md): Agentic coding promises velocity, but without discipline it becomes a shortcut to technical debt. This article sets out which SDLC fundamentals remain non-negotiable (scalability, maintainability, testing and security), which practices must adapt, and which new disciplines, such as prompt management, guardrails for generated code and human-in-the-loop validation, need to be added. - [From LLM to Full Agency](https://dcrewe.com/article/from-llm-to-full-agency_-understanding-the-levels-of-ai-autonomy.md): The term "AI agent" has become so overused it's lost all meaning. This article cuts through the hype by establishing a clear three-level framework for AI autonomy: Single-LLM Features (stateless functions), Workflows (orchestrated processes), and true Agents (goal-driven systems). Learn why precision in AI terminology matters for executives, engineers, and the industry's credibility. - [Always-On AI](https://dcrewe.com/article/always-on-ai-designing-resilient-systems.md): This article discusses the unique challenges of building resilient AI systems and proposes a 5-layer model for ensuring high availability. It emphasizes the need for a hybrid edge/cloud architecture and for balancing the cost of resilience with its business impact. The article provides a checklist for leaders to implement a resilience-first approach. - [LLMs for Intent Classification](https://dcrewe.com/article/llms-for-intent-classification_-why-you-dont-need-a-full-ml-team.md): The rise of large language models (LLMs) has transformed intent classification, making it possible to achieve production-quality results without a full machine learning team or the need to train a model from scratch. The article explains how prompt-driven classification, using zero-shot or few-shot examples, offers a faster, more flexible, and more cost-effective alternative to traditional, data-heavy ML pipelines. It also addresses engineering considerations such as latency, ambiguity, cost, and evaluation, while highlighting how this approach is already being used successfully in various fields. - [Memory, the Missing Link](https://dcrewe.com/article/why-memory-is-the-missing-link-in-ai-agent-design.md): This article argues that memory is the crucial missing piece for the next generation of AI agents, enabling them to be coherent, reliable, and capable of building relationships with users. It breaks down different types of memory essential for effective agents, such as conversational, persona, and episodic memory. The article also explains that proper memory management is key, and it introduces "Agentic RAG" as a superior approach to traditional RAG by using selective, strategic retrieval. Finally, it emphasizes that memory should be treated as a foundational architectural concern for building trustworthy and effective AI agents. ## Pages - [About](https://dcrewe.com/about): Technology leader, consultant and builder focused on what happens when software engineering, artificial intelligence and organisational knowledge start working as one system. - [LinkedIn](https://www.linkedin.com/in/donovan-crewe) - [All articles](https://dcrewe.com/articles) - [Contact](https://dcrewe.com/contact) - [RSS feed](https://dcrewe.com/feed.xml)