Start here

A clear learning plan for someone with AWS and DevOps experience.

You do not need to learn everything at once. Start with the basics, then connect them to your work in Kubernetes, Jenkins, Git, Terraform, AWS cost optimization, and platform operations.

Learning note: This is educational guidance only. If a step, example, or plan does not feel useful for your situation, ignore it and keep the parts that help you move forward.

What is AI?

AI is software that can recognize patterns, understand input, and produce useful output. It is not just a chatbot; it can be used for classification, summarization, advice, and workflow help.

What is an LLM?

An LLM is a large language model. It is trained on huge amounts of data and predicts the most likely next text. It can explain incidents, summarize logs, or help you review infrastructure code.

What is a prompt?

A prompt is the instruction you send to the model. A good prompt gives context, role, task, and expected output. A weak prompt creates weak results.

What is an agent?

An agent is an AI system that follows a goal and uses tools to complete work. It is more than a single answer because it can plan, decide, and act on data.

RAG

RAG means Retrieval-Augmented Generation. The AI searches your internal docs or runbooks first, then uses that information to answer better.

Structured outputs

Instead of free-form text, you can ask the model to return a specific format such as JSON, bullet lists, or a table. This makes AI easier to use in automation.

Tool calling

Tool calling means the AI can invoke functions or APIs. This lets it read logs, query systems, or run actions based on real context.

Why this matters

These tools are what make AI useful for infrastructure and DevOps work, not just for content generation.

Good first project ideas

Terraform code review assistant, AWS cost optimization helper, Jenkins pipeline troubleshooting agent, or Kubernetes ingress issue analyzer.

How to build it

Pick one problem, define the goal, write a prompt, add a workflow, choose tools, and test the output. If it fails, fix the prompt or workflow before adding more complexity.

Example

An AWS cost agent reads Terraform and identifies unused resources, oversized EKS nodes, or old instances that can be right-sized.

Skills

A skill is a reusable instruction set for one job, such as cost review or pipeline troubleshooting. This keeps the agent focused and consistent.

Memory

Memory helps the agent retain useful context across tasks. Use it carefully so the model does not get overloaded with irrelevant history.

MCP

MCP gives AI a standard way to connect to external tools and data sources. This makes integrations cleaner and easier to reuse.

Evaluation

Every agent should be tested on real scenarios. Check whether the answer is correct, safe, and useful before trusting it in production.

How to build an agent

Step-by-step process in simple language

01

Define the goal

Decide what task the agent should solve. Example: review Terraform files and suggest cost savings.

02

Write the prompt

Give the model clear instructions: role, task, constraints, expected output, and what to do when unsure.

03

Design the workflow

Break the work into steps: read input, collect data, analyze, decide, and produce result.

04

Add tools

Use search, API calls, file reading, logs, Terraform files, or scripts to help the agent act on real data.

05

Create a skill

Teach the agent how to behave for a specific job, such as AWS cost review or pipeline troubleshooting.

06

Test and improve

Run it on real examples, review errors, and tighten prompts, workflow logic, and tool limits.

Example agent

Simple AWS cost review agent

Goal: review AWS usage and suggest savings.
Prompt: “You are an AWS cost optimization assistant. Review the input, identify waste, explain risks, and suggest 3 practical savings actions.”
Workflow: read account data → detect oversized resources → compare usage patterns → suggest improvements → output a summary.
Tools: CloudWatch data, cost explorer data, Terraform files, and a cost analysis script.

Multi-agent workflow example

One part does research, another does cleanup, another validates

Research Agent: collects AWS cost data and Terraform drift details.
Analysis Agent: compares usage against expected sizing and flags obvious waste.
Recommendation Agent: creates a short report with risk, savings estimate, and action plan.
This is a common orchestration pattern when one agent handles discovery and the next handles synthesis.