The Emergence of the Context Engineer

In my earlier pieces Frontier AI Isn’t Intelligence — It’s Memory, Beyond Intelligence: The Architecture of Memory, and From Intelligence to Infrastructure: The Rise of Modular AI, I argued that the future of AI will be decided by how we design, maintain, and govern memory. This time, the focus is on the people who make that reliability possible.

Generative AI is adaptive. It can be wrong without hesitation. Without the right context, it does not ask for clarity. It hallucinates. In Beyond Intelligence, I described context rot — the slow drift in coherence and relevance. This is not a syntax error. It is a reasoning failure. Automated, scripted testing rarely detects it.

The Probable–Deterministic Divide

Traditional software is deterministic: the same input gives the same output. Testing verifies that invariance. AI systems, especially long-running agents with memory, are probabilistic. Output can shift with sampling, retrieval state, or memory updates. Enterprise AI will contain both. Deterministic layers will enforce compliance and transaction safety. Probabilistic layers will handle reasoning and adaptation. This is not a case for abandoning automation. It remains essential for deterministic checks and regression. Rather, it means applying the right tools and human skills to the right problems in the AI stack. Automation remains essential for deterministic checks, but human skills are critical for AI reliability.

The Rise of the Context Engineer

The Context Engineer role, described in Beyond Intelligence, designs and governs how prompts, retrieval, and tools interact to sustain coherent behaviour. This is the architecture that prevents agents from operating without the right information. Some organisations are already moving here. Acentra Health has listed AI-QA roles focused on model integrity. Oxford University Press is hiring engineers to shape the “invisible curriculum” of AI tutors. These roles demand technical fluency and the judgement to decide what context an AI actually needs to succeed.

Experienced manual testers bring skills AI reliability demands: adversarial thinking to red-team models for robustness and safety, empathic testing to check relevance and usability, edge-case hunting to stress ambiguous contexts, and context awareness to structure retrieval for intended use. Open-source AI stacks like LangChain, LlamaIndex, and Haystack are already embedding QA-driven workflows into test harnesses, evaluation pipelines, and scenario generation. Many contributors moving into AI reliability come from QA backgrounds.

Signals from the 2024 DORA Report

The 2024 DORA Report analysed more than 39,000 responses worldwide. It found AI use in development correlates with productivity gains and better documentation. It also recorded a 1.5% drop in delivery throughput and a 7.2% drop in stability. but also a 1.5% drop in delivery throughput and a 7.2% drop in stability. Productivity is improving, but predictability suffers. Reliability requires human oversight.

Upskilling for the AI Era

Transitioning a QA professional into AI reliability takes three steps. Recognise the fit: QA is already a foundation for context governance. Upskill in embeddings, retrieval pipelines, evaluation metrics, and explainability. Embed these roles into delivery and governance under titles like Context Engineer or AI Reliability Engineer. AI will scale only as far as the humans who guard its context allow it to. Organizations that fail to equip these professionals risk efficiency gains undercut by unpredictability.

Your most valuable AI reliability engineer might already be in your QA team. The question now is how quickly you can equip them for this critical role.