AI Concepts for Beginners

Imagine that you have hired an exceptionally capable university graduate to be your secretary. They are intelligent, articulate, and able to learn quickly—but this is their first day at your company.

This workplace analogy makes the main parts of an AI system much easier to understand.

The core analogy

Model: the secretary's brain

A model, such as GPT or Claude, is the secretary's brain.

The secretary understands language, reasons about unfamiliar problems, and produces polished writing. However, intelligence alone does not give them access to your company's computers, accounts, files, or internal systems.

A model can think, but it cannot automatically act inside your organization.

Connected apps: accounts and permissions

Connected apps are the accounts and permissions you grant to the secretary.

For example, you might authorize access to company email, messaging, calendars, cloud storage, web search, or another business system. These connections determine which services the secretary can reach and which data or actions are available.

Connection does not mean unlimited access. A well-designed system uses explicit authorization and grants only the permissions required for the task.

Plugins: capability toolboxes

Plugins are capability toolboxes installed in the secretary's workspace.

A plugin can package tools, instructions, application integrations, or other supporting resources. One plugin might help with documents, another with spreadsheets, and another with browser tasks.

Plugins extend what the AI workspace can do, but they are not the intelligence that performs the reasoning.

MCP: the standardized connection interface

MCP, short for Model Context Protocol, is the company's standardized power strip or universal USB interface.

Without a shared standard, every database, application, and tool can expose a different connector. Each integration then requires its own custom adapter. MCP defines a common way for an AI system to discover available capabilities, send requests, and receive results.

MCP answers this question:

How can the AI connect to and communicate with this capability in a consistent way?

MCP is not the database, email service, or business tool. It is the standardized connection layer between the AI system and those capabilities.

Skill: the work manual or SOP

A Skill is the secretary's work manual, playbook, or standard operating procedure.

Connecting the secretary to a database does not teach them which table to query, how to avoid unsafe operations, how to validate the result, or how to write a useful report. A Skill supplies the domain-specific process for completing a type of work correctly and reliably.

A Skill can describe:

• which steps to follow;
• which tools to use;
• which checks to perform;
• what must never be changed or disclosed;
• what a successful result should look like.

MCP explains how to connect. A Skill explains how to do the job well.

Agent: the complete worker

An agent is the secretary as a complete, goal-directed worker.

The agent uses a model to think, follows Skills to choose a reliable process, uses plugins and tools when appropriate, and reaches connected applications through interfaces such as MCP. It can plan multiple steps, inspect intermediate results, recover from some errors, and continue until the requested outcome is complete.

A model provides intelligence. An agent combines that intelligence with instructions, tools, connections, state, and an execution loop.

How the parts work together

Suppose you give the agent this instruction:

Export the three best-selling products from last month's sales database, write an analysis report, and email it to the boss.

The system can handle the request as follows:

1. The agent accepts the goal. It identifies that the task requires data retrieval, analysis, document creation, and email delivery.
2. The Skill supplies the procedure. The agent follows the company's reporting playbook: verify the date range, query sales data safely, rank products, validate totals, write the analysis, and confirm the recipient.
3. Connected apps provide authorized access. The company database and email account are available because the user or administrator granted the required permissions.
4. MCP provides a consistent connection. The agent discovers and calls the database and email capabilities through a standardized interface.
5. Plugins and tools perform specialized operations. A database tool retrieves the records, an analysis tool calculates the ranking, and a document tool prepares the report.
6. The model interprets the information. It explains why the top products performed well and turns the numbers into a clear narrative.
7. The agent completes the workflow. It checks the report and sends the finished result to the authorized email recipient.

The outcome is not merely a chat response. The requested work is completed from beginning to end.

The relationship at a glance

How this maps to Nexts

Nexts is the workspace where the user and the AI worker meet. It brings conversations, files, meetings, models, Skills, plugins, and agent execution into one desktop environment.

Nexts is designed around a local-first boundary:

• private desktop content stays local whenever possible;
• the local data plane owns conversations, files, meeting content, agent execution, and product state;
• external tools and services require explicit connections and authorization;
• the account service handles identity, subscriptions, entitlements, device binding, and usage metadata—not private desktop content.

This separation matters. Giving an AI worker more capability should not mean giving every remote service access to the user's private workspace.

One-sentence summary

The model thinks, the Skill teaches the method, the plugin packages capabilities, MCP standardizes the connection, connected apps grant authorized access, and the agent brings everything together to finish the job.