Listen while you learn

AI Agents: A Practical Guide

AI Agents, MCP, ACP, A2A — they're all new terms that can feel abstract at first. The truth is, this space is still unfolding and there's plenty more to learn. What I've done here is capture my take so far in one place, with the aim of breaking things down to what really matters.
By René Bossa LinkedIn

AI Agents

Intelligent assistants that transform enterprise operations
By René Bossa
↓
Next up: Definition
01 — Definition

What are AI Agents?

AI agents are intelligent assistants that go beyond what people have previously experienced with chatbots or traditional automation. Instead of following rigid scripts, they can understand context, make decisions within defined boundaries, and complete tasks from start to finish, often by working together with other agents or systems.
Scope
They can be applied across the organisation–answering questions, processing requests, or coordinating activities between departments.
How it works
Agents interpret instructions, decide the best next step, and take action by connecting to the right tools (APIs) or data.
What it isn't
They aren't uncontrolled AI. Agents operate under company rules and permissions, ensuring their actions are safe and compliant.
↓
Next up: Value
02 — Value

Why enterprises should care

AI agents deliver measurable impact through efficiency, consistency, and enhanced employee and customer experiences.
Efficiency at scale
Agents take on repetitive or time-consuming tasks, freeing employees to focus on higher-value work.
Consistency & reliability
Processes are followed the same way every time, reducing errors and ensuring compliance.
Improved experiences
Employees and customers get quicker, smoother outcomes, without unnecessary delays or hand-offs.
↓
Next up: Vision
03 — Vision

Where this is heading

The future brings specialised agents working seamlessly together across business functions and industries.
Collaborating agents
Different specialised agents–HR, finance, IT, customer service–working together to handle complex workflows.
Seamlessly integrated
Agents becoming part of the everyday systems employees already use, reducing friction and complexity.
Industry-specific solutions
Ready-made agents tailored to sectors like banking, healthcare, or retail, built to meet regulatory requirements.
↓
Next up: Scenario
04 — Scenario

Employee onboarding

A new hire joins the company. Instead of HR manually coordinating every step, intelligent agents handle the process:
HR agent
creates the employee record.
IT agent
sets up access and equipment.
Learning agent
assigns onboarding and compliance training.
Payroll agent
confirms salary and benefits.
HR agent
provides a progress update back to the manager.

With AI agents (what this means)

Every step is coordinated automatically.
The new hire has a smooth, timely experience.
HR focuses on people, not admin.

Without AI agents (what it looks like)

HR must chase each department manually.
Steps are often missed or delayed.
Onboarding feels slow and inconsistent.

Model Context Protocol

A new standard for AI-enterprise integration
By René Bossa
↓
Next up: Definition
01 — Definition

What is MCP?

Model Context Protocol (MCP) is a standard that lets applications, tools, and systems share context with large language models in a consistent way.
Scope
A common format for describing data, tools and requests so AI systems can understand and use them.
How it works
Defines predictable request/response shapes, discovery of available tools, and permission-aware context passing.
What it isn't
Not a model or product; it doesn't replace your security or data platforms—it standardises how AI talks to them.
↓
Next up: Value
02 — Value

Why enterprises should care

MCP changes day-to-day delivery by cutting integration effort, improving control, and avoiding lock-in.
Speed & cost
Reusable connectors and patterns reduce custom build time and maintenance.
Control & assurance
Clear rules on what context can flow, who can access it, and an auditable trail.
Vendor flexibility
Portable integrations across models and platforms keep you free to switch as needs evolve.
↓
Next up: Vision
03 — Vision

Where this is heading

Standards turn pilots into platforms and enable broader collaboration.
Interoperable agent networks
Multiple agents from different vendors coordinating through a shared protocol.
Certified marketplace & portable policies
Pre-built MCP connectors and policy packs you can adopt, review, and govern.
Industry profiles
Sector-specific patterns (e.g., banking, healthcare) that align MCP with regulations and common workflows.
↓
Next up: Scenario
04 — Scenario

Damaged order replacement

Customer: "My parcel arrived damaged." Your AI agent resolves it by calling specific MongoDB MCP tools, no direct database access from the agent.
Locate order
`find` on `orders` with `{ orderNumber, email }` to return minimal fields (status, items, address).
Check eligibility
`aggregate` across `orders` + `returns_policy` to confirm it's within the replacement window.
Open a case
`insert-one` into `service_cases` and capture the `caseId`.
Queue replacement
`update-one` on `orders` to set `status: "replacement_pending"` and attach `caseId`.
Confirm to customer
"Case SC-10492 created; replacement arriving in 3–5 working days."

With MCP (what this means)

Only approved actions (`find`, `aggregate`, `insert-one`, `update-one`) are exposed; least-privilege and easy to audit.
Config-level controls (read-only modes, allow/deny lists) reduce risk from prompt mistakes.
Central logging; swapping systems or changing schemas is contained to the MCP layer.

Without MCP (what it would look like)

The AI agent or bespoke middleware needs direct DB credentials and custom code for each step.
Inconsistent guardrails and scattered logs; harder to prove who did what.
Tighter coupling to your database–schema/vendor changes trigger rewrites and increase break risk.

Communication Protocols

A new standard for agent-to-agent collaboration
By René Bossa
↓
Next up: Definition
01 — Definition

What are communication protocols (ACP & A2A)?

Communication protocols like ACP (Agent Communication Protocol) and A2A (Agent-to-Agent Protocol) are standards that let AI agents talk to each other in a common language. They ensure agents can exchange information, delegate tasks, and collaborate, regardless of who built them or where they run.
Scope
A shared rulebook so agents can work together smoothly, rather than needing custom connections each time.
How it works
Agents describe their capabilities, share information in consistent formats, and follow agreed rules for secure hand-offs.
What it isn't
These protocols don't make decisions or perform tasks themselves–they simply provide the framework for clear communication.
↓
Next up: Value
02 — Value

Why enterprises should care

Communication protocols transform collaboration by enabling faster onboarding, better control, and vendor flexibility.
Speed & cost savings
Standard connections mean faster onboarding of new agents without expensive custom integration.
Control & compliance
Clear rules define what information can be shared, who can access it, and provide an auditable record of activity.
Flexibility
Agents from different vendors can work together, helping avoid lock-in and allowing the business to evolve without major rewiring.
↓
Next up: Vision
03 — Vision

Where this is heading

Standards enable seamless collaboration across business functions and vendor ecosystems.
Connected agent ecosystems
HR, finance, IT, and customer service agents all collaborating seamlessly within one business.
Trusted marketplaces
Pre-built, certified agents available "off the shelf" that can plug into existing systems with confidence.
Industry alignment
Communication patterns tailored for sectors like banking, healthcare, or retail, ensuring agents follow regulations and common workflows.
↓
Next up: Scenario
04 — Scenario

Customer refund handling

A customer requests a refund. Instead of a single system handling everything, different agents step in through a communication protocol:
Customer agent
captures the request.
Finance agent
validates payment details.
Compliance agent
checks refund policy rules.
Logistics agent
confirms the return of goods.
Customer agent
provides a unified update back to the customer.

With ACP / A2A (what this means)

Each agent knows what the others can do and passes tasks securely.
The business has a clear record of who did what.
Agents from different vendors work together without extra engineering.

Without ACP / A2A (what it would look like)

Each connection requires custom coding.
Security and logging vary by system, making compliance difficult.
Changes to vendors or systems cause costly delays and rewrites.
📱

Not quite mobile-ready…

Apparently building a website that behaves the same everywhere is harder than it looks...

Right now, I can only support desktop – but I’m working on making it mobile-friendly too, so do bear with me.

For now, enjoy the desktop version while I figure out how to make this experience just as engaging on your phones too.

– René