A client walks into a law office and slides a tablet across the desk. Instead of the usual hour spent recounting the basic facts of a dispute, the lawyer sees a structured summary. The client has already spent the morning interacting with an AI that guided them through a specific legal triage process, identifying the critical points of contention and organizing the necessary evidence. The professional no longer wastes the first thirty minutes of a consultation on data collection. They dive immediately into high-level strategy because the AI has already performed the heavy lifting of the initial intake.
The Architecture of k-sajja-agents
This shift in professional interaction is the driving force behind k-sajja-agents, an open-source AI agent skill registry designed to codify the tacit knowledge of Korean professionals. Rather than treating AI as a general-purpose chatbot, k-sajja-agents treats professional expertise as a set of discrete, deployable skills. The project operates as a structured repository where professional knowledge is categorized by occupation, with further sub-directories for specific contributors or firms.
The technical foundation of the registry rests on two primary files. The first is `SKILL.md`, which serves as the operational blueprint. This file defines exactly how a specific task should be performed, outlining the logic, the sequence of questions, and the criteria for evaluation. The second is `PROFILE.md`, which acts as the professional's digital identity, containing their specialization, professional biography, contact information, and direct booking links.
In practice, these skills are tailored to the nuances of different fields. A lawyer might contribute a skill specifically for reviewing startup equity agreements or real estate contracts, embedding their unique perspective on risk management into the AI's logic. A physician might design a workflow that systematically organizes symptom progression, medication history, and diagnostic test queries. A tax accountant might define a precise sequence for classifying evidentiary documents and generating a pre-filing checklist. Crucially, the registry maintains a strict boundary between process and authority. The AI is designed to handle the workflow, but it is never granted the power to make final decisions, such as issuing a medical prescription, determining a final tax liability, or filing a legal motion.
To lower the barrier to entry for non-technical experts, the project provides a meta-skill called `sajja-skill-creator`. This tool allows professionals to build their own `SKILL.md` files through a simple question-and-answer interface, removing the need for coding knowledge. Once the professional has structured their workflow through this guided process, they can submit their skill to the registry via a Pull Request, integrating their expertise into the open-source ecosystem.
From Knowledge Monopoly to Workflow Interfaces
For decades, the economic value of professional services was built on the foundation of information asymmetry. Experts maintained their competitive edge by keeping their proprietary methodologies hidden, charging premiums for access to a black box of specialized knowledge. However, k-sajja-agents signals a fundamental reversal of this model. In the age of agentic AI, the most powerful portfolio is no longer a list of past clients, but a public, functional demonstration of how one thinks and works.
By publishing an Agent Skill—a set of instructions that defines how an AI should execute a task—the professional transforms their expertise into a transparent interface. This creates a new kind of lead generation funnel. Potential clients can experience the expert's methodology through the AI before ever booking a call. When a client sees that the AI, powered by a specific professional's skill, can accurately triage their problem, it builds a level of trust and validation that a traditional marketing brochure or blog post cannot achieve.
This transition effectively automates the low-value, repetitive portions of professional work, pushing the human expert further up the value chain. The AI handles the intake and organization, while the human intervenes exactly where the AI reaches its cognitive limit. This is not merely an efficiency gain; it is a complete redesign of the client acquisition path, moving from passive advertising to active, tool-based engagement.
This shift also redefines the relationship between developers and domain experts. In previous iterations of legal-tech or med-tech, developers interviewed experts to build a closed system. Now, the experts themselves distribute their workflows as open-source assets. As frontier models from OpenAI, Gemini, Anthropic, and Grok evolve to better call external tools and follow complex instructions, these skills can function as standardized APIs for professional knowledge. The expertise is no longer trapped in a PDF or a human brain; it is a machine-readable instruction set that any compatible LLM can invoke to provide high-fidelity professional guidance.
The value of the modern professional is migrating away from the possession of knowledge and toward the design of the process used to solve a problem.
