Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallThe skill-driven enterprise is an architectural proposal, published by Manovikas Muduganti on DEV Community on September 16, 2026, that separates reusable business know-how from the AI agents that apply it. In this model, a business skill holds the instructions, decision logic, required context, expected outputs, and quality criteria for a task. An agent is assembled to execute that task, and a surrounding platform called the harness decides what the agent is allowed to touch. The author presents this as a design direction, not an established enterprise standard or a proven operating model.
The core idea: separate what the business knows from who executes it
Most enterprise agent projects today bind knowledge, tools, permissions and behavior into one long-lived agent for each function. The proposal argues that this makes every agent a bottleneck: the know-how for an invoice dispute or a customer refund lives inside one agent’s prompt and configuration, where it is hard to reuse, test or govern. The alternative the author describes uses a shared layer of skills and a per-task assembly of agents on top of it. The framing the article uses is a chain: Intent → Skill → Agent → Governed Execution → Verified Outcome.
As an Amazon Associate I earn from qualifying purchases.
The building blocks of the model
Business skills
A skill is a reusable package that describes how capabilities are applied to a meaningful business task. According to the article, a skill can contain:
- instructions for the work and the business knowledge it depends on
- decision logic and the context the task requires
- the expected outputs
- the tools or capabilities it needs
- the policies that apply
- criteria for judging whether the work was done well
The distinction the article draws is between a skill and a tool. A tool supplies a single capability, such as searching documents or retrieving a customer record. A skill decides how those capabilities are combined and judged for a particular business outcome.
#1 Best Overall
The skills marketplace
The proposal includes a place where teams publish, discover, reuse, version, test and improve skills. The article describes this as a proposed component. It does not identify an existing marketplace or claim that one is in general use, so readers should treat it as a requirement to build or buy rather than a product to adopt.
The agent harness
The harness is the surrounding platform that does the coordination. In the author’s description it interprets intent, selects a skill, checks prerequisites and user access, assigns the capabilities the task is allowed to use, applies policy, routes approval requirements, executes the work and evaluates the result. The central governance principle is that the agent should not determine its own permissions. Access decisions sit with the harness, not with the model that is acting.
MCP and enterprise capabilities
The article presents the Model Context Protocol (MCP) as a standardized way for agents to reach enterprise systems and tools. Its role in the model is narrow: MCP provides capability access, and the harness decides which capabilities an agent receives. The article does not show that MCP by itself handles identity management, policy enforcement or risk governance. Those functions are assigned to the harness and to the organization’s own controls.
Task-specific agents
Instead of keeping a permanent agent for every business function, the author suggests assembling an agent when a task needs one. The components are an agent template, a skill, the relevant context, the MCP capabilities the task requires, applicable policies and evaluators. This is presented as a direction for the design, and it is not demonstrated in the cited article with deployment data.
The skill lifecycle
The proposal treats skills as products that change over time. Its lifecycle runs in five stages, and each one has a specific check:
- Create. Write the skill’s instructions, context requirements, expected outputs and criteria.
- Test. Run structural checks and permission checks, then evaluate the skill against realistic scenarios.
- Publish. Release a tested version to the marketplace so other teams can discover and reuse it.
- Observe. Track how agents running the skill behave in production.
- Improve. Revise the skill based on what was observed, and version the change.
The article asks evaluators to check five things about an agent working from a skill: whether it follows the skill, whether it uses suitable information, whether it stays within its permissions, whether it escalates when it should and whether its output is useful.
Rank #4
How a request moves through the harness
The article’s more detailed sequence has nine stages. Reading them in order shows where the separation between know-how and execution is enforced:
- Intent: the business request is stated in plain terms.
- Identity: the requester is identified and their access is established.
- Context: the information relevant to the request is gathered.
- Skill: the harness selects the matching business skill.
- Prerequisites: the harness checks that the skill’s required context, tools and conditions are present.
- Agent: an agent is assembled from the template, skill, context and permitted capabilities.
- Policy: the policies attached to the skill and the organization are applied, including approval requirements.
- Execution: the agent performs the work with the capabilities it was granted.
- Evaluation: the result is checked against the skill’s criteria.
The useful question for an implementation team is not whether each stage exists, but which component owns each decision. In this model, the agent executes, while the harness and the skill’s policies decide.
Best Value
Where governance fits: the NIST AI Risk Management Framework
The article does not connect its design to a formal governance framework, so the closest external reference is the NIST AI Risk Management Framework (AI RMF). NIST describes it as voluntary guidance for incorporating trustworthiness into the design, development, use and evaluation of AI systems. It is general guidance, not a validation of the skill-driven architecture. Its core is organized into four functions, which map onto the proposal as follows.
| NIST AI RMF function | What it asks an organization to do | Where it applies in a skill-driven setup |
|---|---|---|
| Govern | Define policies, accountabilities and roles, including human-AI oversight. NIST’s core explicitly calls for defining and differentiating roles and responsibilities for human-AI configurations. | Who owns each skill, who approves changes to its policies, and who can grant the harness new capabilities. |
| Map | Document intended purposes, context, users, assumptions and potential impacts before deciding whether to proceed. | Which business tasks a skill covers, which users it serves, and which tasks it should never be assembled for. |
| Measure | Evaluate security, resilience and other relevant risks, test before deployment and regularly in operation, and document the methods and results. | Scenario evaluations of skills, permission-check results, and production observation of agents running them. |
| Manage | Prioritize assessed risks, decide whether the system meets its objectives, and plan response and ongoing monitoring. | Deciding when a skill version is retired, when escalation rules change, and how incidents feed back into the Improve stage. |
NIST’s own AI RMF documentation notes that AI RMF 1.0 is being revised. Before citing a specific version in a policy or procurement document, confirm the current status on the NIST AI Resource Center’s AI RMF Core page. NIST’s January 26, 2023 announcement described the framework’s consensus development as involving more than 240 organizations across private industry, academia, civil society and government. That figure describes how the framework was developed. It does not measure adoption or demonstrated effectiveness. In the announcement, NIST Director Laurie E. Locascio said: “The AI Risk Management Framework can help companies and other organizations in any sector and any size to jump-start or enhance their AI risk management approaches.” That is NIST’s statement of the framework’s intended usefulness, not an independent evaluation of any particular architecture. The announcement is available from NIST’s January 2023 news release.
Deciding between permanent agents and task-specific assembly
The proposal raises a practical question that many teams will face: how many agents should we build? It suggests a different question as the better test: how easily can we assemble the right agent for the work? The following axes are the ones the article’s design implies. The cited material does not provide head-to-head performance data for either approach, so each row is a question to answer inside your own organization rather than a result.
| Decision axis | Question to ask | What a good answer looks like |
|---|---|---|
| Reuse | Are the business rules for a task used by more than one workflow? | The rules live in one skill with a named owner, not copied into several agent prompts. |
| Permissions | Who decides what an agent can access, and can the agent change that? | The harness assigns capabilities; the agent cannot grant itself new ones. |
| Prerequisites | What happens when required context or a tool is missing? | The task stops or escalates at the prerequisite check instead of proceeding on partial information. |
| Testing | How are evaluation criteria written and kept current? | Each skill has scenario tests and criteria that are versioned with it. |
| Observability | Can you tell which skill version an agent ran, and what it did? | Every execution is traceable to a skill version, a capability set and an outcome. |
| Maintenance | How much effort goes into templates and skills compared with long-lived agents? | The team can estimate the upkeep of each approach from its own pilot work. |
What is established and what is not
- Established: the proposal is a single author’s architecture, published on September 16, 2026, that separates reusable skills from the agents that execute them.
- Established: NIST’s AI RMF is voluntary guidance organized into Govern, Map, Measure and Manage, and NIST has stated that it is being revised.
- Not established: adoption rates, productivity gains, cost reductions or scalability improvements for business skills, skill marketplaces, agent harnesses or task-specific agents.
- Not established: that an existing marketplace or product implements this model at scale.
- Not established: that MCP alone provides identity management, policy enforcement or risk governance.
Any benefit claimed for this architecture should be treated as a hypothesis until an organization measures it on its own workflows, with its own permission model and its own evaluation criteria.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
