For the past year, the consulting world has been saturated with talk of "Agent strategies." Yet, this focus on the execution layer has largely ignored the more critical foundation: Skills.
Barry and Mahesh from Anthropic delivered a particularly incisive critique of this trend: instead of endlessly building new, bespoke Agents, the strategic imperative is to build, compound, and reuse Skills around a single, general-purpose Agent. In essence, this means packaging executable consulting methodologies into structured, callable modules, allowing AI to finally develop true domain-specific craftsmanship.
This analysis will deconstruct how consulting firms can operationalize this Skill-centric approach.
This approach codifies the tacit knowledge of your best performers—how they conduct due diligence, structure industry models, or execute interview scripts. The strategic value of this "Skills-based" approach to AI adoption lies in two core principles: low organizational friction and high accessibility.
First, the implementation is minimally invasive. Forget the monolithic, high-risk "company-wide Agent platform" that requires years of integration and organizational upheaval. Instead, you can begin with a single department or business line. The initial task is not to build a new system, but to operationalize existing assets. Convert your current Excel templates, interview guides, analysis scripts, and presentation structures into a repository of executable Skills. The most compelling value proposition for large enterprises is not migrating everyone to yet another new system; it's teaching an AI Agent to navigate and execute tasks within their existing, often idiosyncratic, internal software and workflows. This path offers immediate value with minimal disruption.
Second, this framework is accessible enough for non-technical experts to contribute directly. Anthropic observed that many of the most valuable Skills are being developed by professionals in finance, legal, and HR. These experts are not engineers; they are codifying their daily operational playbooks using simple markdown and lightweight scripting, creating precise instructions for an Agent to follow. In a professional services firm, the most valuable assets are the project managers and niche subject-matter experts. If they can author Skills without learning to code, you are no longer constrained by the bandwidth of a few engineers. You are mobilizing the entire intellectual capital of the firm.
Here is an example of a Skill Anthropic developed internally for creating presentations:
The concept of "Skills" offers a more engineered, systematic solution to leveraging AI. A particularly potent idea, especially for professional services, is to reframe Skills as an evolving capability knowledge base, jointly maintained by the organization and its AI Agents. As teams interact with models, correcting their outputs and enriching them with institutional knowledge, this repository of skills deepens. Crucially, every Agent within the organization can then access this shared, compounding intelligence. This is precisely the architecture we are building at Epsilla. We see this "capability knowledge base" as a Semantic Graph—a living, dynamic repository where organizational intelligence is stored not as static documents, but as interconnected, executable capabilities.
Consider the evolutionary path of a single Skill.
It begins simply: an "Industry Report Generation Skill." The initial version might be a straightforward markdown template outlining the structure of a 30-page industry analysis—specifying the required data, charts, and case studies for each chapter, perhaps augmented with a few macros or Python scripts for basic data processing.
Next, a project team finds this template insufficient for a "Go-to-Market Strategy" engagement. They fork the original Skill, adding modules for overseas regulatory analysis, currency risk assessment, and local competitive landscapes. As this new version is used across projects, it organically accumulates a rich layer of practical knowledge—common pitfalls to avoid and best practices from highly-rated engagements.
Eventually, the knowledge management function intervenes, treating these Skills with the same rigor as software assets. They introduce version control, maintain changelogs, and write automated tests for critical Skills (e.g., given a set of inputs, does the Agent-generated outline meet specific quality standards?). This aligns with Anthropic's own thinking about evaluating, versioning, and managing dependencies for Skills to make Agent behavior more predictable and reliable.
This approach unlocks a new paradigm for enterprise knowledge management. Knowledge is no longer a collection of searchable documents; it becomes a library of executable units of capability. Every course correction on a project, every framework refinement, is no longer just tacit experience in an employee's mind. It is codified back into the Skill, becoming the default starting point for the next team. Humans and Agents are collaboratively building the organization's proprietary operating system. At Epsilla, we believe the Semantic Graph is the ultimate repository for these executable capabilities—the central nervous system where an organization's intelligence lives, breathes, and evolves. Our Agent-as-a-Service platform is designed to consume these Skills directly from the graph, turning static knowledge into dynamic action.
From our experience driving AI transformation in professional services, the inflection point is never the adoption of a new model. It's the moment a team realizes their scattered PowerPoint roadmaps and interview scripts can be codified into a persistent, improvable Skill for an AI Agent. In that instant, the knowledge base transforms from a passive "document warehouse" into an active "capability platform."
If your organization is investing in knowledge management but finds the output is underutilized and static, it's time to ask a critical question: Should you shift focus from "writing documents" to "engineering Skills"? This is more than a semantic shift; it's a fundamental change in how you build and scale institutional intelligence. It's about creating a unified context management system where your company's core competencies are not just stored, but are alive and executable. This is the future we are building.
The Obsolescence of the Slide Deck: Why Your Firm Must Shift from Presentations to Executable Skills
If your organization has invested heavily in knowledge management, yet the resulting repository feels inert—"difficult to use, and no one reads it"—it's time for a fundamental re-evaluation. The problem isn't the content; it's the container. The strategic imperative is to shift from creating static presentations to engineering dynamic, executable "Skills."
The Consultant's Dilemma: A Microcosm of Enterprise Inefficiency
Consider the classic kickoff meeting in any high-stakes consulting project. The client invariably states, "Our industry is incredibly complex. Can you even understand it?" The consulting partner smiles and replies, "Give us two weeks. We'll start with a comprehensive industry scan."
The reality behind this exchange is a brutal race against time. A team of highly paid analysts must go from zero to a 60% proficiency level—just enough to be effective—in a matter of days. The traditional workflow is a frantic scramble: consuming industry analyses, scanning research papers, and tracking regulatory changes, all to be synthesized and "packaged into a story" by a human with strong structuring abilities.
A generalist AI Agent, despite its power, faces the same core problem. It's akin to the "300 IQ math genius who doesn't know the 2025 tax code." You cannot afford for an Agent to derive a complex regulatory framework from first principles; the computational and temporal cost is simply too high.
The true value of a Skill is that it codifies the very process of "how to rapidly enter a new industry," transforming it from an ad-hoc, artisanal effort into a reusable, procedural asset.
For the individual consultant, this model introduces a profound benefit: the externalization and standardization of their learning curve. The traditional, often haphazard apprenticeship under a project manager is replaced by a structured, AI-guided execution path. The Agent, equipped with a specific industry Skill, essentially runs the consultant through a real-world training simulation. It assigns daily readings, poses targeted questions, and provides micro-assignments. The consultant's feedback and work product are then used to refine the Agent's understanding, which is written back into the core Skill.
The tangible experience is one of symbiotic growth. The consultant observes their "AI colleague" becoming demonstrably smarter each day, ceasing to repeat the same cognitive errors about the industry. Simultaneously, their own professional development becomes more visible, systematic, and accelerated.
From a career and organizational standpoint, this approach to "industry-entry Skills" has a subtle but critical implication. When a consultant eventually leaves the firm, they take their accumulated industry insights and experience with them. However, the core methodologies, analytical frameworks, templates, and learning pathways remain behind, codified as a durable corporate asset. This intellectual property, now a permanent part of the firm's arsenal, can be used to train the next generation of consultants. This creates a cleaner, more transparent value exchange between the individual and the firm.
This is where the architecture of knowledge management becomes paramount. These methodologies and playbooks are not just files in a shared drive; they are the foundational elements of a firm's collective intelligence. The most effective repository for this is not a folder structure but a Semantic Graph, like the one we've built at Epsilla. This graph doesn't just store the "Skills"; it understands the relationships between them, creating a unified context that powers every Agent interaction. It is the firm's true long-term memory.
A Pragmatic Roadmap for Implementing Agent Skills
This vision often prompts a critical question from managing partners: "This sounds transformative, but where do we begin? We can't afford a year-long R&D project, but we also don't want to be stuck in a perpetual proof-of-concept."
Based on our experience deploying AI transformations within professional services, we advocate for a starting point that is both lightweight and grounded in real-world workflows.
Step 1: Target a High-Frequency, Standardizable Scenario. Resist the urge to solve every problem at once. Begin by identifying a task that is both common and follows a relatively fixed process. Prime candidates include:
- Initial industry landscape scans
- Competitive intelligence reports
- Structuring raw interview transcripts
- Generating first drafts of proposals These are weekly, if not daily, activities in most firms, making them ideal candidates for codification into a first-generation Skill. Step 2: Codify the Playbook Before Touching the Technology. The initial focus should be on knowledge capture, not model selection. The design philosophy here is to make Skill creation accessible to non-technologists. A senior consultant should be tasked with documenting the process exactly as they would write a project playbook. This document must clearly articulate the steps, decision criteria, quality standards, and frequently used templates. Only after this domain expertise is captured should AI engineers or developers translate the automatable portions into scripts or tools. The objective is not to write complex code, but to create a Skill that faithfully mirrors the firm's proven method of execution. This documented playbook becomes a foundational asset, ingested into Epsilla's Semantic Graph to provide structured context for the Agent. Step 3: Iterate in a Small Team on Live Projects. Do not attempt a firm-wide rollout from version one. The initial implementation should be treated as a closed beta. Deploy the Skill with a single, agile project team and empower them to use, critique, and directly modify it. The feedback loop must be tight and immediate. Encourage the team to log their frustrations and improvements directly back into the Skill's documentation. This iterative process treats your firm's intellectual property with the same rigor as software development, incorporating version control, change tracking, and even automated evaluation. This requires a new role—someone responsible for curating and maintaining this portfolio of capabilities. Far from being an administrative burden, this function evolves into the core asset management of the 21st-century consulting firm. Every iteration doesn't just improve a document; it enriches the firm's central Semantic Graph, compounding the value of its intellectual capital with every project executed. Anthropic’s framing of Skills evolution mirrors the discipline of software development, complete with version control, behavioral tracking, and even automated evaluation. For professional services and consulting firms, the implication is profound. The responsibility of "maintaining this set of industry capabilities" transforms from an administrative overhead into a core function of strategic asset management.
Step 4: Institutionalize Skills, Don't Just Build Demos
Once a firm develops 5 to 10 robust Skills—spanning key industries or critical functions—the focus must shift from creation to governance. This is the inflection point where most initiatives either scale or stagnate. The critical questions become operational: How do you manage permissions to ensure certain Skills remain siloed within specific practice groups? How do you create a unified access point within your existing knowledge infrastructure? How do you assign clear ownership and define a maintenance cadence for each Skill?
Failure at this stage is predictable and costly. It leads directly to the digital graveyard of abandoned knowledge bases, where valuable assets go to die from neglect. Success, however, creates something entirely new: a dynamic, executable repository of proprietary methodologies, not a static archive of unsearchable legacy documents. This is the difference between a liability and a compounding asset.
This is precisely where a system designed for complex, interconnected knowledge becomes non-negotiable. Traditional knowledge bases fail because they are passive repositories. To manage a living ecosystem of Skills, you need a system that understands relationships, dependencies, and context. At Epsilla, we architected our Semantic Graph for this exact purpose. It serves as the ultimate repository for these executable Skills, managing not just the content but the intricate web of permissions, ownership, and versioning. It is the foundation for turning scattered methodologies into a coherent, manageable asset pool.
Our work with leading professional services firms validates this phased approach. We typically begin with a tightly scoped use case, helping a team package methodologies from disparate sources like SharePoint or internal wikis into structured, executable Skills. These Skills are then orchestrated by an Agent, powered by our Agent-as-a-Service platform, which draws its context and capabilities directly from the Semantic Graph. We prove the model within a single business unit or industry vertical before methodically expanding across the organization.
This brings us back to the provocative mandate: “Don’t build agents, build skills instead.”
For the consulting world, the message is even more precise: Stop fixating on building the perfect, monolithic "consulting LLM." The more urgent and strategic question is this: What percentage of your firm’s unique industry frameworks and project playbooks has been converted into executable, version-controlled Skills?
If the answer is close to zero, you are not behind. You are perfectly positioned at the starting line of a fundamental transformation in how intellectual capital is managed and deployed.