For decades, learning and development (L&D) leaders have looked for solutions to Benjamin Bloom’s 2-sigma problem: the uncomfortable truth that while one-on-one mentoring is the most effective form of teaching, it’s often too costly and labor-intensive to scale across an organization.

Artificial intelligence (AI), when used well, is emerging as a means of scaling a high-touch apprenticeship experience to an extended learner audience. But in order to create a premium, hyper-personalized learning strategy, you need to understand context engineering.

Consider this parallel: You’ve just walked into a local ice cream shop that boasts a homemade recipe. Your expectations are high. You aren’t there for a pre-packaged, mass-produced ice cream sandwich that tastes like cardboard; you want a high-quality treat served exactly the way you like it. To meet your expectations, the shop needs three things: the homemade ice cream, someone who knows how to wield a scooper and the right vessel (bowl, cone, etc.) to hold the ice cream.

Too often, the corporate world settles for an AI experience much like that mass-produced ice cream sandwich. We push a button, receive a generic response and wonder why the output doesn’t meet learner expectations. Let’s dig into how to fix this generic learner experience.

The AI Context Architect and the Multi-Layered Framework

Context engineering is not just writing a good prompt; it’s building the multi-layered framework an AI tool uses to understand the learner’s world. This process involves the deliberate shaping of a foundational model to understand the specific organizational context, job role and task to produce a precise and accurate answer.

To build this multi-layered framework, you’ll need to start with a knowledge base. To create it, you point a large language model (LLM) to the additional context and background information it needs to deliver tailored, relevant and accurate information to your learners. Returning to our ice cream shop, the knowledge base is like the rich cream every flavor is based upon.

The Technical Foundation: Knowledge Bases and RAG

A knowledge base’s technical architecture consists of three key components: the foundational LLM, context engineering and comprehensive prompt design. Here’s a high-level look at each layer:

• Foundational LLMs: Google Gemini, OpenAI’s ChatGPT or Anthropic’s Claude are trained on massive volumes of text covering a wide variety of information. Though powerful, they tend to produce generic responses when used “off the shelf.” That’s because a public LLM lacks insight and knowledge about your organizational culture, history, processes and language that define your brand. If learners ask a generic LLM a question related to their daily work, they’ll get a generic output that may be irrelevant at best and misleading at worst.
• Context engineering: To move beyond a generic output, L&D leaders and teams must work with IT, engineering, stakeholder and subject-matter expert (SME) teams to build an organizational knowledge base that will contain the answers your learners and organization need.

The knowledge base is your organization’s internal source of truth. It is a centralized repository of your proprietary playbooks, technical documentation, cultural values and ethics: your own proprietary “recipe.” This repository primes the LLM with knowledge specific to your organization, learners and tasks, which directs it to deliver information that’s relevant and useful.

Retrieval-augmented generation (RAG) connects the foundational LLM to your knowledge base. Think of it as the server who takes the customer’s order, retrieves the ice cream and then serves their order.

RAG is built upon a system of code that fetches only the most relevant facts from your knowledge base in real time. Instead of scanning a 500-page manual for every response, RAG identifies the exact page that applies to the learner’s current inquiry. Thus, RAG ensures the output remains accurate and grounded in the learner’s reality.

• Comprehensive prompt design: Context architecture is already part of the L&D leader’s job description. To succeed as context architects, we must know how to structure our instructions for our LLM to define its role, goals and constraints so that it knows exactly how to interact with and respond to the learner.

Training the AI Trainer: Comprehensive Prompt Design in Context

We typically initiate new projects by gathering information on the learner personas, the most appropriate learning modality and the organization’s regulations and brand voice. As context architects, we not only need to identify this information but also train our LLM on it to build a truly personalized learning experience.

The necessary information includes three main categories:

1.     The Individual: Personalizing the Learning Experience

By providing the LLM with data on the learner’s role, seniority and past performance, L&D equips the LLM with the tools to adapt to who is on the other side of the screen. This data allows it to frame the training in a way that is immediately relevant to the learner’s specific learning needs and interests.

In addition, if, through context engineering, we are able to capture “curiosity” data — information about where a learner pauses, the questions they ask or what they skip — we gain deeper insight into how people actually learn by tracking the paths they choose and the questions they have along the way, which enables us to shape future training accordingly.

2.     The Environment: Mapping the Organizational Ecosystem

This data is all about the organizational ecosystem the individual works within: whether they work in a casual remote environment, a formal office environment or anything in between. Understanding the environment ensures that the LLM delivers a relevant answer that reflects the reality of the learner’s workplace.

Imagine two managers taking the same course on conflict resolution. In a remote-first tech startup, the LLM might coach the manager to resolve issues via video call, per the organizational culture of speed and flexibility. In a legacy firm with a strict hierarchical culture, the LLM might instead suggest a formal in-person meeting and provide a template for human resources (HR) documentation.

3.     The Constraints: Establishing AI Guardrails and Brand Standards

These are the non-negotiable guardrails the LLM must work within. Every organization has “must-know” compliance points, safety regulations or a specific brand voice that cannot be compromised. By including these rules, you ensure that the LLM delivers output that is aligned with these standards and avoids the legal or cultural pitfalls of generic LLM output.

For example, a financial services firm might create a constraint requiring the LLM to always use specific language to disclaim investment risks, preventing a learner from unintentionally generating unauthorized financial advice. Conversely, a legacy luxury brand might impose a constraint that forbids the LLM from using slang terms or informal contractions to ensure a consistently sophisticated tone.

The Human-AI Partnership: Augmenting Intelligence

Becoming a context architect doesn’t mean L&D leaders and professionals are training our replacements. In fact, we’re augmenting our human intelligence — and that of our teams and experts — and extending its reach to a much wider audience than was previously possible.

As always, no one works alone. Impactful AI-powered learning requires a blend of specialized humans working in harmony. These roles include the:

• Learning experience designer (LXD): The architects who define the objectives and lead the context engineering process to ensure the LLM approaches learners in the right role, with the right tone and feedback, to teach them the defined learning and performance objectives.
• Developer and engineer: The technical specialists who handle technical execution and ensure data integrations are robust and secure. One of their vital roles is building a secure intermediary between your organizational knowledge base and the foundational LLM to ensure that no sensitive or proprietary information is “leaked” to the public.
• User experience (UI) specialist: The advocates for the learner and creators of intuitive interactions. They make sure the technology is seamless and never impedes or overshadows the learning experience.

L&D Leadership in the Age of AI

As AI technology continues to evolve, part of the value we bring as L&D leaders lies in our ability to remain in the driver’s seat, working across teams to refine and curate our organizational knowledge base, monitor and supervise AI outputs, and ensure that every interaction and decision is grounded in human intentionality.

With the partnership of AI, new solutions to the 2-sigma problem are finally within reach, allowing us to recapture the one-to-one training experiences of the apprenticeship era. As we continue to advocate for both our people and our business, our perspectives as L&D leaders will be more needed to coach, mentor and teach “the one” at the scale of “the many.”