There are two kinds of intelligence on Earth: human intelligence and artificial (or, rather, alien) intelligence. Both kinds of intelligence operate in the same way – they follow an old coders' rule – "garbage in -> garbage out". This is a pretty simple rule: good output requires good input. More specifically, intelligence cannot generate smart and useful material unless it has received comprehensive and detailed information.

Two months ago Tobi Lutke – CEO of Shopify, multibillionaire and car racer – posted a tweet that was viewed by 1.7 million people:

"I really like the term "context engineering" over prompt engineering. It describes the core skill better: the art of providing all the context for the task to be plausibly solvable by the LLM."

The following week, Tobi's opinion was validated by one of AI industry superstars - Andrej Karpathy - with a longer tweet, specific and self-explanatory:

"+1 for "context engineering" over "prompt engineering". People associate prompts with short task descriptions you'd give an LLM in your day-to-day use. [But]… context engineering is the delicate art and science of filling the context window with just the right information for the next step. Science because doing this right involves task descriptions and explanations, few shot examples, RAG, related (possibly multimodal) data, tools, state and history, compacting... Too little or of the wrong form and the LLM doesn't have the right context for optimal performance..."

In a nutshell, context engineering is all about carefully crafting and managing the environment and the inputs around an AI system so it can perform optimally. It's like giving the AI the right "frame" to understand problems and respond usefully, rather than just hoping it'll figure out everything on its own.

Aren't we lucky that human intelligence has been roaming planet Earth for many thousands of years and we have some experience in providing context to people, coming from the most relevant field of management. In corporate-speak these two procedures are called "recruitment" and "onboarding" and HR departments are supposed to be good at this.

The recruitment usually involves getting acquainted with a person's education and understanding their previous work experience and projects. The onboarding process includes sharing corporate policies and procedures, best practices and operation manuals.

Let's think how this human experience can be applied to AI systems. The parallels between human context setting and AI context engineering are striking and offer us a proven framework to work with.

When we hire humans, we rely heavily on their existing educational foundation. We assume they come with basic knowledge about their field, general problem-solving skills, and cultural understanding. Similarly, for AI agents, we need to establish this foundational layer - what we might call the "education layer." This involves feeding the AI comprehensive knowledge bases, domain-specific information, and fundamental principles that will serve as its baseline understanding.

The second layer in human context setting comes from previous work experience. We value candidates who have worked in similar roles or industries because they bring tested approaches and learned patterns. For AI context engineering, this translates to what we call "best practices integration." We need to provide AI agents with documented successful approaches from other organizations, industry standards, and proven methodologies. This layer gives the AI a repository of "what works" in similar situations.

The third and most crucial layer is company-specific knowledge. During onboarding, we share our unique culture, internal processes, specific tools, and organizational quirks that make our workplace different from others. For AI agents, this becomes the "organizational knowledge database" - all the internal documents, company-specific workflows, historical decisions, and institutional memory that makes the AI truly useful within our particular context.

But here's where it gets interesting - and where we might actually need to do better than traditional HR practices. While humans can fill in gaps through intuition and social learning, AI agents need explicit context engineering. We can't rely on their ability to "pick things up" through office conversations or learn unwritten rules through observation.

This means we need to formalize what has often been informal in human management. We need to document the undocumented, make explicit the implicit assumptions, and create structured context-building processes that go far beyond what we typically do for human employees. In essence, working with AI agents might actually make us better at managing humans too, by forcing us to be more deliberate and comprehensive in how we share context and knowledge.