Artificial Wisdom

Context engineering promises to make artificial intelligence genuinely intelligent

A new discipline is emerging that could transform how machines think, and how humans work alongside them

In the summer of 2025, as artificial intelligence (AI) grew increasingly sophisticated, a curious limitation emerged: digital amnesia. Ask ChatGPT to draft a follow-up email, and it politely requests a reminder about who "John" is and yesterday’s discussion. Seek assistance with a complex project, and you find yourself repeating background details endlessly, akin to briefing a helpful but forgetful intern each morning.

The solution lies not in crafting bigger digital brains, but rather in superior memory management. Enter "context engineering," a discipline so new its foundational principles emerged from a viral tweet barely three months ago, yet so fundamental it may determine corporate success in the AI era.

The twenty-minute memory conundrum

To grasp context engineering, consider the constraints facing even the most advanced AI. Today’s large language models, while powerful, retain limited information at once. Claude, a leading AI assistant, "remembers" roughly 150,000 words. GPT-4 manages about half as many. Beyond these thresholds, information disappears, creating a working-memory constraint oddly reminiscent of human limitations.

Unlike humans, who compensate with notes, intuition, or brief memory refreshers, AI systems have historically lacked such flexibility. Context engineering addresses this, strategically deciding what an AI should "remember" at any moment. As Andrej Karpathy, formerly of Tesla and OpenAI, notes: "Context engineering is the delicate art and science of filling the context window with just the right information for the next step."

From craft to science

The discipline crystallised around an insight from Tobi Lütke, Shopify’s chief executive, who argued that "context engineering" better described the practice than the common "prompt engineering." Rather than simply refining queries, effective AI use now involves curating precisely the right contextual information—timely, relevant, and succinct.

Take Manus, an AI coding assistant used by millions. Rather than treating each interaction as isolated, Manus employs "externalised memory"—a structured file system enabling information retention across sessions. When tackling complex programming tasks, it generates and methodically updates to-do lists, referring back to earlier decisions. This transforms AI from mere tools into knowledgeable collaborators.

The business implications are profound. Early adopters report astonishing productivity gains: security bottlenecks cut by 60%, deployment cycles shortened by 30%, and hundreds of hours annually reclaimed from compliance work. One financial services firm saw document processing accuracy leap by 18 percentage points after implementing context engineering.

Memory as competitive edge

Context engineering challenges traditional assumptions about AI advancement, shifting the focus from model size and capability to superior information management. The key to creating genuinely intelligent AI systems lies in managing not only how they think but what they think about.

Companies mastering this discipline extract significantly more value from existing AI models. Context-aware systems offer truly intelligent experiences...recalling user preferences, understanding business intricacies, and building coherently upon past interactions.

Early pioneers are already reaping substantial rewards. Shopify leverages context-aware AI for personalised customer support. Anthropic employs multi-agent research systems vastly superior to single-agent counterparts. These innovations provide not incremental improvements but entirely new user experiences.

The substrate imperative

Context engineering requires a robust infrastructure—"universal substrates" for representing and sharing knowledge. Just as HTTP underpinned the early internet, standardised knowledge-sharing protocols might become indispensable for future AI interactions.

My personal contribution in this area is the Knowledge Substrate Engine (KSE), a commercial solution designed to manage and represent knowledge effectively, underpinning context-aware AI applications.

Companies establishing these foundational technologies could dominate as database firms did in the 1980s or cloud providers in the 2000s....supplying essential services upon which entire ecosystems depend.

Revisiting the attention economy

Context engineering also reframes the central challenge of digital abundance: information overload. Previously, technology companies competed fiercely for human attention. Now, context engineering suggests the next wave of competition will centre on managing attention effectively, helping individuals and businesses focus on the most critical tasks.

In enterprises, AI increasingly resembles personalised assistants rather than general-purpose tools, deeply aware of specific business contexts, remembering prior decisions, and providing finely tuned advice.

Risks and realities

Yet context engineering brings challenges. Foremost are privacy concerns: AI that remembers everything inevitably accesses sensitive information, raising stakes for security breaches or misuse.

Further risks include dependency and skill erosion. As AI handles complex contexts adeptly, humans risk losing critical cognitive skills. Like GPS navigation’s convenience diminishing spatial awareness, context-aware AI may erode human capacity to independently manage complexity.

Moreover, context engineering remains as much empirical art as systematic science. Practitioners joke about "stochastic gradient descent," highlighting the trial-and-error nature of their craft. Early advantages may thus be fleeting until best practices become widespread.

Towards artificial wisdom

Despite these limitations, context engineering introduces a genuinely novel prospect: AI systems that feel truly intelligent rather than merely capable. The distinction is qualitative...a knowledgeable advisor rather than a mere search engine; a trusted colleague rather than a sophisticated calculator.

For businesses, the implication is clear: context engineering is not simply a technical curiosity but a foundational capability that will decisively shape competitive positions in AI-driven markets. Early mastery promises systems that genuinely enhance human intelligence, remembering essential details, forgetting trivialities, and offering precisely the right information exactly when needed.

The age of artificial intelligence may be here, but artificial wisdom is only beginning.

Context engineering shows that the path forward involves teaching machines the quintessentially human skill of discerning what deserves attention. In an age saturated with information, that may indeed represent intelligence of the highest order.

Welcome to the era of Artificial Wisdom.