By Sadhana | Sadhana’s Journal — Learn. Grow. Inspire.
Artificial Intelligence is changing how we learn, work, create, and solve problems. From asking an AI assistant to explain a difficult concept to generating software code, analysing data, creating images, and automating repetitive tasks, AI is becoming a useful part of everyday life.
But what exactly is Artificial Intelligence? Are all AI tools the same? What is the difference between Generative AI and Reactive AI? Which tools are free, which require payment, and how can a beginner start using them effectively?
If you have these questions, this guide is for you.
Whether you are a student, software developer, business professional, software tester, content creator, or Oracle HCM techno-functional consultant, this article will help you understand AI from the basics and explore practical ways to use it.
You do not need a technical background to get started. We will begin with simple explanations and gradually move into AI tools, prompting, automation, Codex, and responsible AI use.
Table of Contents
- What is Artificial Intelligence?
- How does AI work?
- AI, Machine Learning, Deep Learning, and LLMs
- Types of AI explained
- Reactive AI versus Generative AI
- Generative AI, Predictive AI, and Agentic AI
- Free, paid, open-source, and proprietary AI
- Popular AI tools and how to choose them
- AI for students
- AI for developers
- AI for business professionals
- AI in entertainment and creative work
- AI for software testers and QA professionals
- AI for Oracle HCM techno-functional consultants
- Prompt engineering: How to communicate with AI
- How to automate repetitive tasks
- Understanding Codex and AI coding agents
- Limitations, risks, and responsible AI
- A 30-day AI learning roadmap
- Frequently asked questions
- Final thoughts
1. What Is Artificial Intelligence?
Artificial Intelligence, commonly called AI, is a field of computing that develops systems capable of performing tasks that normally require human-like abilities such as recognising patterns, understanding language, making predictions, solving problems, or generating content.
Think about the technology you already use.
- Your email application identifies possible spam messages.
- A navigation application estimates travel time and suggests routes.
- A shopping website recommends products based on your activity.
- A voice assistant responds to spoken requests.
- An AI chatbot explains a topic or helps draft an email.
- An image-generation tool creates a picture from a written description.
These applications use different AI techniques. Some classify information, some predict outcomes, and others generate new content.
A simple example
Imagine you want to write an email requesting leave from your manager.
Without AI, you might think about the wording, draft the message, and revise it yourself.
With a generative AI assistant, you could write:
“Draft a polite email requesting one day of leave for a personal appointment. Keep it professional and under 100 words.”
The AI produces a draft that you can review, correct, and personalise.
The important point is that AI assists with the task. You remain responsible for deciding whether the output is appropriate.
Is AI the same as a robot?
No. A robot is a physical machine that can perform actions in the real world. AI is a collection of computational techniques.
A robot may use AI to identify objects or navigate its surroundings, but many AI systems exist entirely as software. A chatbot, spam filter, or code assistant does not need a physical body.
In simple terms: AI is not one application or one machine. It is a broad field of technology that helps computers perform a variety of tasks.
2. How Does AI Work?
Different AI systems work differently, but a simplified explanation is that many AI systems use data, algorithms, and computational models to produce an output.
Consider a system that identifies whether an email is spam.
- The system receives an email.
- It examines features such as words, links, sender information, and patterns.
- A model or set of rules evaluates the information.
- The system assigns a classification, such as spam or not spam.
- The email application uses that result to decide where the message should go.
A generative AI assistant follows a different process.
- You provide a prompt.
- The application sends your request and relevant context to an AI model.
- The model processes the input and generates an output.
- The application presents the response.
- You review the result and ask follow-up questions if needed.
Many modern language models generate text by predicting likely next tokens based on the input and patterns learned during training. A token may be a word, part of a word, punctuation, or another unit of text.
This does not mean the model always knows whether its response is factually correct. A fluent answer can still contain mistakes.
Training versus inference
Two important terms are worth understanding.
Training: The process through which a model’s parameters are adjusted using data and a learning method.
Inference: Using a trained model to produce a prediction, classification, or generated output.
For example, a language model may be trained before you ever use it. When you submit a question and receive an answer, the model is performing inference.
Not every AI system uses the same training method, and some applications also use additional information retrieval, tools, rules, or calculations.
Does AI think like a human?
AI can perform tasks that resemble aspects of human reasoning, but that does not establish that it thinks, understands, or experiences the world in the same way a person does.
It is better to evaluate AI by what it can reliably do, where it fails, and how well it performs on the task you need.
3. AI, Machine Learning, Deep Learning, and Large Language Models
These terms are related, but they are not interchangeable.
Artificial Intelligence (AI)
The broad field of building systems that perform tasks associated with capabilities such as perception, reasoning, prediction, language processing, and decision-making.
Machine Learning (ML)
A branch of AI in which systems learn patterns from data rather than relying exclusively on explicitly programmed rules.
Example: A model learns from historical transactions to identify patterns associated with potentially fraudulent activity.
Deep Learning
A type of machine learning that uses neural networks with multiple processing layers to learn complex patterns.
It is widely used in image recognition, speech processing, language models, and other applications.
Example: A deep-learning model identifies objects in a photograph.
Large Language Models (LLMs)
LLMs are models trained on large amounts of language data to process and generate text. Many can also work with code, and some support images or other types of input.
Examples of applications built around language models include chat assistants, document summarisation tools, and coding assistants.
How do these concepts relate?
A simplified relationship is:
- AI is the broad field.
- Machine learning is one approach within AI.
- Deep learning is a family of machine-learning techniques.
- Many modern language models use deep-learning architectures.
Not every AI application uses machine learning, and not every machine-learning model is a language model.
4. Types of AI Explained
There is no single classification that covers every type of AI. AI can be grouped according to its capabilities, its method of operation, or the task it performs.
Two classifications are especially useful for beginners.
4.1 Classification by capability
Narrow AI
Narrow AI is designed to perform particular tasks or a limited range of tasks. It includes systems for translation, recommendations, image recognition, fraud detection, and conversational assistance.
Most AI systems used in everyday life today fall into this broad category, including highly capable general-purpose assistants.
Artificial General Intelligence (AGI)
AGI generally refers to a hypothetical or debated level of AI that could perform a broad range of intellectual tasks with generality comparable to human capabilities.
There is no universally accepted test for AGI, and claims about whether a system has achieved it depend on the definition being used. Current AI capabilities should not automatically be treated as proof that AGI has been achieved.
Artificial Superintelligence (ASI)
ASI is a hypothetical form of AI that would substantially exceed human capabilities across many important intellectual domains. It is not an established everyday technology.
4.2 Classification by operation or task
Other useful categories include:
- Reactive AI
- Limited-memory AI
- Predictive AI
- Generative AI
- Recommendation systems
- Agentic AI
These categories can overlap. For example, a modern AI assistant can generate text, use retrieved information, and invoke tools as part of a multistep workflow.
5. Reactive AI Versus Generative AI
This distinction is important because the terms describe different aspects of AI.
What is Reactive AI?
Reactive AI is a classic category of AI system that responds to the current situation without maintaining an internal history of past experiences to guide its decisions.
A traditional chess program that evaluates the current board position and selects a move is a familiar example.
A reactive system can still be sophisticated. “Reactive” does not mean unintelligent or necessarily based only on simple rules.
What is Generative AI?
Generative AI creates new content based on patterns learned during training and the input it receives.
It can generate:
- Text and explanations
- Images and illustrations
- Computer code
- Music and audio
- Video
- Summaries and other documents
For example, you can ask a generative AI assistant to draft a blog introduction or create a Python function.
Comparison
| Feature | Reactive AI | Generative AI |
|---|---|---|
| Main purpose | Respond to the current situation | Generate new content |
| Typical output | A move, classification, or action | Text, images, audio, video, or code |
| Historical context | Classic reactive systems do not maintain an internal history of previous situations | Context and memory depend on the model and application |
| Example | A classic chess program evaluating a board | An AI assistant drafting a blog article |
| Main concern | Quality of the decision or response | Quality, accuracy, originality, and safety of generated content |
These are not mutually exclusive categories. Reactive AI describes a classic way of operating, whereas generative AI describes a capability. A modern system can generate content and also react to new inputs.
What is limited-memory AI?
Limited-memory AI uses information from recent inputs or stored historical data to support its output. Many machine-learning systems use historical data during prediction, and some AI applications maintain conversational context or application memory.
However, a model that can use the previous messages in a conversation does not necessarily learn permanently from those messages. The application’s memory, the model’s training process, and the conversation context are different things.
6. Generative AI, Predictive AI, and Agentic AI
Generative AI: Create something new
Use generative AI when you need a first draft, an explanation, a summary, an image concept, or a code example.
Example: “Create five beginner-friendly practice questions about TypeScript arrays.”
Predictive AI: Estimate what may happen
Predictive AI uses data and models to estimate an outcome, category, or future value.
Examples include demand forecasting, estimated delivery times, customer churn predictions, and risk scoring.
A prediction is an estimate, not a guarantee.
Recommendation AI: Suggest relevant options
Recommendation systems rank or suggest items based on information such as preferences, behaviour, similarity, and context.
Examples include recommended products, films, music, and articles.
Agentic AI: Work through a sequence of tasks
An AI agent can use a model together with tools, instructions, and a workflow to pursue a goal through multiple steps.
For example, a coding agent might inspect a repository, identify a bug, edit code, run tests, and propose changes for review.
An agent is not automatically reliable or fully autonomous. Its actual capabilities depend on the tools, permissions, safeguards, and environment it has been given.
Traditional automation versus AI-powered automation
Traditional automation follows predefined rules and steps.
For example, a script may rename every file in a folder according to a fixed naming pattern.
AI-powered automation can handle tasks involving less-structured information, such as summarising incoming requests or categorising support tickets. It may be combined with conventional automation for predictable actions.
A good workflow often combines both: AI handles interpretation, while deterministic rules handle checks and actions that must be consistent.
7. Free, Paid, Open-Source, and Proprietary AI
Not all AI tools use the same business model or make their models available in the same way.
Free AI tools
Free plans let users try AI without paying a subscription. They may have limits on the number of requests, access to advanced models, file uploads, image generation, or research features.
Free does not necessarily mean open-source or unrestricted.
Paid AI tools
Paid plans may provide higher usage limits, more capable models, additional features, faster access, collaboration, or business controls.
Paying for an AI tool does not guarantee that every answer will be correct.
Open-source AI
Open-source software makes its source code available under a licence that grants specified rights. In AI, it is important to examine which components are actually open: the application code, model implementation, training code, model weights, and datasets may have different licences.
Open-weight AI
An open-weight model makes its trained model weights available for use under a particular licence. Some open-weight models can be run on your own computer or organisation’s infrastructure.
Open-weight does not automatically mean unrestricted commercial use, full transparency about training data, or compliance with every open-source definition.
Proprietary AI
A proprietary model is controlled by its provider and may be accessible through a hosted application or API without making its underlying weights publicly available.
This can be convenient because you do not need to manage the infrastructure, but you should consider subscription costs, usage limits, privacy terms, and provider dependency.
Which should you choose?
- Choose a free hosted tool when you are learning or have occasional tasks.
- Consider a paid plan if you regularly hit limits or need specific features.
- Explore open-weight models if you need local execution, experimentation, or greater infrastructure control.
- Consider business or enterprise offerings when your organisation needs administrative controls, contractual protections, and approved data handling.
Always check the current licence, data policies, and usage terms before using a model for commercial or confidential work.
8. Popular AI Tools and How to Choose Them
The AI landscape changes rapidly. Tools add features, change usage limits, and introduce new plans. The following examples are a starting point, not a permanent ranking.
General-purpose AI assistants
ChatGPT — Useful for explanations, writing, brainstorming, data analysis, coding assistance, and supported multimodal tasks.
Google Gemini — Useful for conversational assistance, writing, research, and workflows that integrate with supported Google products.
Claude — Useful for writing, analysing documents, coding assistance, and working through complex instructions.
Microsoft Copilot — Useful for general assistance and, with eligible Microsoft 365 plans, AI features integrated into supported productivity applications.
Coding and development tools
OpenAI Codex — A coding agent for software-development tasks, with capabilities depending on the product experience and available tools.
GitHub Copilot — An AI coding assistant integrated into supported development environments and GitHub workflows.
Cursor — An AI-focused code editor with coding assistance and agent features.
Research and information discovery
Perplexity — An AI-powered search and answer product designed to help users explore information and source material.
ChatGPT, Gemini, and Claude can also support research tasks, depending on the features available in the selected plan.
For factual research, prefer responses with relevant sources and verify important claims against the original material.
Design, images, and creative work
Canva — Helps create presentations, social media graphics, documents, and other visual materials, with AI features available under varying plan limits.
Adobe Firefly — Provides generative creative tools for supported image, video, and audio workflows.
Exploring open models
Hugging Face — A platform for discovering models, datasets, and AI development resources.
Ollama — Helps users run supported models locally and explore model-based workflows on compatible hardware.
Running a model locally may improve control over where prompts are processed, but privacy depends on the complete setup, connected services, and how data is handled.
How much do AI tools cost?
Some tools offer free plans alongside paid subscriptions. For example:
- ChatGPT offers free access and paid plans with different capabilities and limits.
- Claude offers a free plan and paid individual plans.
- GitHub Copilot offers a free tier and paid developer plans.
- Cursor offers a free entry-level option and paid plans.
- Canva and Adobe Firefly offer free access with usage limits, alongside paid creative plans.
- Ollama supports local model execution, while its hosted services and usage options have separate terms.
Pricing varies by country, taxes, billing period, usage, and product. Check the provider’s official pricing page before subscribing. An application subscription and API usage can also be billed separately.
Official pricing pages:
- ChatGPT: https://openai.com/chatgpt/pricing/
- Claude: https://claude.com/pricing
- Google AI: https://one.google.com/about/google-ai-plans/
- Microsoft Copilot: https://www.microsoft.com/en-in/copilot/
- OpenAI Codex: https://openai.com/codex/
- GitHub Copilot: https://github.com/features/copilot/plans
- Cursor: https://cursor.com/pricing
- Perplexity: https://www.perplexity.ai/
- Canva: https://www.canva.com/en_in/pricing/
- Adobe Firefly: https://www.adobe.com/in/products/firefly/plans.html
- Hugging Face: https://huggingface.co/
- Ollama: https://ollama.com/
How should a beginner select an AI tool?
Start with the task rather than the brand.
Ask yourself:
- What do I want to achieve?
- Can a free plan handle the task?
- Does the tool support the files, formats, or applications I use?
- Can I verify its output?
- What are its privacy and data-retention terms?
- Will a paid plan save enough time or provide enough value to justify its cost?
You do not need to subscribe to several AI tools immediately. Learn one tool well, understand its limitations, and expand only when a real need arises.
9. AI for Students
Students can use AI as a learning assistant, provided they use it to understand material rather than simply copy answers.
Useful applications
- Explain difficult topics in simple language.
- Generate practice quizzes and flashcards.
- Summarise notes and organise revision plans.
- Explain mistakes in practice problems.
- Practise a new language through conversation.
- Brainstorm project ideas and improve presentation structure.
- Ask for alternative explanations when the first explanation is unclear.
Example prompt
“Act as a patient tutor. Explain the difference between variables and arrays in TypeScript using simple examples. Then give me five practice exercises without solutions. Wait for me to answer before giving feedback.”
This prompt creates an interactive learning experience.
What students should avoid
Do not submit AI-generated work as your own when academic rules prohibit it. Verify facts, check calculations, and follow your institution’s rules for AI use.
AI can help you learn, but it cannot replace the understanding you develop by solving problems yourself.
10. AI for Developers
Developers can use AI throughout the software-development lifecycle.
Common use cases
Code generation: Produce an initial implementation from a clear requirement.
Debugging: Explain an error message and suggest possible causes.
Code explanation: Understand unfamiliar code or a complex function.
Refactoring: Improve readability while preserving behaviour.
Testing: Draft unit tests, edge cases, and test data.
Documentation: Generate draft README files, comments, API examples, and technical explanations.
Learning: Explore a new language, library, framework, or programming concept.
Example prompt
“Act as a TypeScript mentor. Write a function that accepts an array of employee objects and returns employees whose status is active. Define the interface, handle an empty array, explain the code, and provide five test cases. Do not use any type unless necessary.”
The developer should still review the code, run tests, check security implications, and verify compatibility with the project.
AI-generated code can contain defects, outdated APIs, insecure patterns, and incorrect assumptions. Treat it as a proposed change, not a verified solution.
11. AI for Business Professionals
AI can help with routine communication, document preparation, planning, and analysis.
Examples
- Draft professional emails.
- Summarise meeting notes.
- Convert rough notes into structured reports.
- Create presentation outlines.
- Compare options against stated criteria.
- Generate frequently asked questions.
- Organise project tasks.
- Help interpret data when supported by appropriate tools.
Example prompt
“Turn the following meeting notes into a concise professional summary. Include decisions made, open questions, action items, owners, and due dates. If any owner or date is missing, mark it as ‘Not specified’ instead of guessing.”
This instruction is useful because it tells the AI what to include and what not to invent.
Business caution
Never upload confidential contracts, employee information, financial records, or customer data to an unapproved AI service. Follow your organisation’s policies and use approved tools for business information.
12. AI in Entertainment and Creative Work
AI is changing how people create and experience entertainment.
Writing and storytelling
AI can brainstorm characters, develop story outlines, suggest titles, or help revise a draft.
Music and audio
Supported tools can generate musical ideas, sound effects, voice content, or assist with audio editing.
Images and video
Generative tools can create illustrations, concept art, backgrounds, storyboards, and video clips.
Games and interactive experiences
AI can assist with character dialogue, development workflows, testing, and procedural content generation.
Example prompt for a visual concept
“Create a visual concept for a peaceful mountain road trip in Andhra Pradesh during light drizzle. Include misty green hills, winding roads, and a small roadside tea stall. Use a realistic travel-photography style and a wide landscape composition.”
For commercial work, check the relevant tool’s licence, rights, and terms. Do not assume that generated content is automatically free of copyright, trademark, likeness, or other legal concerns.
13. AI for Software Testers and QA Professionals
AI can assist testers with test design, documentation, analysis, and automation support.
Practical use cases
- Generate test scenarios from requirements.
- Identify missing acceptance criteria.
- Suggest positive, negative, and boundary-value tests.
- Draft API test cases and assertions.
- Create sample test data.
- Summarise defect reports.
- Suggest regression-test coverage.
- Explain accessibility testing concepts.
- Draft Gherkin scenarios.
- Help analyse logs and error messages.
Example prompt
“Act as a senior functional QA engineer. Review the requirement below and create test scenarios covering positive, negative, boundary, validation, and permission-related cases. Include scenario ID, preconditions, steps, expected results, and priority. Do not invent business rules; list any assumptions separately.”
API testing example
“Given the following API endpoint and approved sample request, propose Postman test cases for successful responses, invalid input, missing required fields, authentication failures, and unexpected response formats. Provide example assertions, but clearly mark any status codes or rules that need confirmation from the API specification.”
AI can speed up preparation, but the tester must validate the scenarios against actual requirements and the system under test.
14. AI for Oracle HCM Techno-Functional Consultants
Enterprise implementation consultants often work across business requirements, application configuration, data, reports, integrations, and testing. AI can help with many documentation-heavy and repetitive activities.
Requirement analysis
Ask AI to summarise business requirements, identify ambiguities, and prepare clarification questions.
Example: “Review these non-confidential recruiting requirements. Map each requirement to a proposed process area, identify missing information, and create questions for the business owner. Do not assume that a feature is supported without verification.”
Configuration documentation
AI can help draft setup checklists, configuration notes, and implementation documentation. Confirm every setup task against the relevant Oracle documentation and the client’s design.
HCM Data Loader (HDL)
AI can explain HDL concepts, help interpret non-sensitive error messages, and suggest validation checks for sample files.
Never assume that generated HDL files are ready for production. Validate business keys, dependencies, security, and data quality in a controlled environment.
OTBI and BI Publisher
AI can help turn reporting requirements into draft specifications, explain SQL concepts, and suggest report test cases.
Verify data models, subject areas, joins, security rules, and performance in the actual environment.
HCM Extracts and Fast Formulas
AI can explain concepts, help interpret errors, and draft starting points for formulas or extract requirements. Validate syntax, available database items, effective dates, and business logic in the relevant Oracle environment.
REST APIs and integrations
AI can explain payload structures, draft test scenarios, and suggest error-handling checks. Verify endpoints, authentication, permissions, and supported operations against official documentation.
Recruiting Cloud example
For an Oracle Recruiting Cloud implementation, AI could help create a checklist for requisition approval, candidate progression, interview scheduling, offer approval, and candidate communication.
However, the consultant must confirm the actual workflow configuration, roles, approval rules, and customer-specific requirements.
The key principle: AI can accelerate analysis and documentation, but it should not replace functional expertise, environment validation, or approved implementation processes.
15. Prompt Engineering: How to Communicate with AI
A prompt is the instruction or input you give an AI system.
Prompt engineering means designing and refining those instructions to improve the relevance, usefulness, and consistency of the output.
You do not need complicated language. Clear instructions are often more useful than long, vague requests.
A practical prompt formula
Try this structure:
Role + Task + Context + Constraints + Output format
- Role: What perspective should the AI take?
- Task: What exactly should it do?
- Context: What information does it need?
- Constraints: What must it include or avoid?
- Output format: How should the answer be organised?
Weak prompt versus strong prompt
Weak prompt:
“Write about AI.”
This gives the model little direction.
Improved prompt:
“Act as a beginner-friendly technology writer. Write an introduction to Artificial Intelligence for readers with no technical background. Explain AI using three everyday examples, avoid unexplained jargon, use short paragraphs, and finish with a transition to the different types of AI.”
The second prompt explains the audience, task, scope, style, and desired ending.
Useful prompting techniques
Zero-shot prompting: Ask for a task without providing examples.
“Explain what an API is in simple terms.”
Few-shot prompting: Provide examples of the desired output.
“Classify each defect as Critical, Major, or Minor using these examples and definitions. Then classify the new defect.”
Step-by-step task decomposition: Break a complex task into smaller deliverables.
“First summarise the requirement. Next identify ambiguities. Then draft test scenarios. Finally, create a coverage matrix.”
Output constraints: Specify a format, length, tone, or required fields.
“Return the results as a table with columns for Scenario, Preconditions, Steps, Expected Result, and Priority.”
Grounded prompting: Provide trusted source material and ask the AI to base its answer on that material.
“Use only the policy text below. If the answer is not supported by the text, say ‘Not specified in the provided policy.'”
Improve the answer through follow-up prompts
You do not need to get everything right in your first request.
Useful follow-ups include:
- “Explain this more simply.”
- “Give me a practical example.”
- “Identify any assumptions in your answer.”
- “Compare the alternatives in a table.”
- “Check for missing edge cases.”
- “Ask me one question at a time.”
- “Separate verified facts from suggestions.”
Common prompting mistakes
Avoid instructions that are too vague, contradictory, or missing important context.
Do not expect the AI to know your internal project requirements, private documents, or organisation-specific rules unless you provide approved context.
Do not treat a confident answer as proof of correctness. For important tasks, request sources, assumptions, and verification steps.
16. How to Automate Repetitive Tasks with AI
One of the most practical uses of AI is reducing the effort spent on repeated tasks.
The goal is not to automate everything. It is to identify tasks where AI can save time without creating unacceptable risk.
Step 1: Identify repetitive work
List tasks you perform frequently, such as:
- Writing similar emails.
- Preparing meeting summaries.
- Creating recurring test scenarios.
- Drafting status reports.
- Formatting documents.
- Summarising non-sensitive feedback.
- Creating social media captions.
Step 2: Break the task into steps
For example, a weekly status report might involve collecting updates, identifying completed tasks, recording blockers, and writing a summary.
Decide which steps require human judgement and which are repetitive.
Step 3: Create a reusable prompt
Build a template with placeholders for the information that changes.
Example reusable prompt:
“Prepare a weekly project update using the notes below. Organise it into Completed, In Progress, Blockers, Risks, and Next Steps. Use concise professional language. Do not invent dates, owners, or progress percentages. Mark missing information as ‘Not provided.'”
Step 4: Test the workflow
Try the prompt with several representative examples. Check whether it consistently produces the expected format and whether important information is omitted or invented.
Step 5: Add automation where appropriate
Depending on the task, you may use templates, scripts, approved integrations, workflow tools, or AI agents.
For example, an approved workflow might collect non-sensitive feedback, summarise recurring themes, and prepare a draft report for review.
Step 6: Measure the result
Compare the time spent before and after introducing AI. Also evaluate accuracy, rework, consistency, and risk.
A faster process is not necessarily a better process if it creates more errors.
A simple example: QA test preparation
Without AI: A tester manually converts a requirement into test scenarios, formats the cases, and checks coverage.
With AI assistance: The tester provides an approved requirement, asks AI for draft scenarios in a predefined template, reviews the output, and validates the cases against the application.
AI assists with the first draft. The tester remains accountable for coverage and correctness.
17. Understanding Codex and AI Coding Agents
Codex is an example of an AI coding agent designed to help with software-development tasks. Product capabilities depend on the specific Codex experience and its available tools.
A general-purpose AI assistant can explain code or draft a function in a conversation. A coding agent may also be able to work with project files, inspect a codebase, modify files, and run supported development commands.
What can a coding agent help with?
- Explain unfamiliar repositories.
- Implement a clearly defined change.
- Draft unit tests.
- Identify potential bugs.
- Refactor code.
- Update documentation.
- Run tests in an available environment.
- Suggest a fix for a failing test.
Example task instruction
“Inspect the existing TypeScript project and explain its folder structure. Do not modify any files. Identify the main entry point, test framework, and available scripts. If something cannot be established from the repository, state that explicitly.”
This is a good first task because it asks the agent to investigate without making changes.
After you understand the repository, you might ask:
“Implement input validation for the employee creation function. First inspect the existing conventions and tests. Make the smallest necessary change, add tests for valid input and missing required fields, run the relevant test suite, and report any tests that could not be run.”
Safety when using coding agents
Start with a clean version-control state, understand the proposed changes, and review the diff before accepting it. Use a safe test environment and grant only the permissions required for the task.
Do not give an agent unrestricted access to production systems, secrets, or confidential data. Commands and file changes should be reviewed, especially when they can delete data, install packages, change permissions, or access external services.
AI coding agents can make mistakes even when they report that a task is complete. Passing tests provide useful evidence, but they do not prove that all behaviour is correct.
18. Limitations, Risks, and Responsible AI
AI is useful, but responsible use requires understanding its limitations.
Hallucinations
An AI model may generate plausible but incorrect information, references, calculations, or code.
What to do: Verify important facts against reliable sources and test generated code.
Bias
AI outputs can reflect biases present in training data or system design.
What to do: Review outputs carefully, especially for decisions affecting people.
Privacy and confidentiality
Information submitted to an AI service may be processed according to that provider’s terms and settings.
What to do: Follow organisational policy, use approved services, and avoid submitting confidential or personal information without authorisation.
Copyright and ownership
The legal treatment of AI-generated material depends on the jurisdiction, the source material, the tool’s terms, and how the content is used.
What to do: Check applicable licences and policies before publishing or using generated material commercially.
Over-reliance
Relying on AI without understanding the task can create errors that are difficult to detect.
What to do: Use AI to assist your thinking, not to avoid learning or professional review.
Security
Generated code or instructions can contain insecure patterns. AI agents can also make harmful changes if given excessive permissions.
What to do: Apply normal secure-development practices, use least-privilege access, review changes, and test in controlled environments.
A useful rule is to treat AI output as a draft or recommendation until it has been verified to the level required by the task.
19. A 30-Day AI Learning Roadmap
You do not need to learn every AI tool at once. A structured plan can help you build confidence gradually.
Week 1: Understand the basics
- Learn AI, machine learning, deep learning, and LLM terminology.
- Understand Generative AI, Predictive AI, Reactive AI, and AI agents.
- Try one general-purpose AI assistant.
- Ask it to explain a topic you already know and compare the response with your own knowledge.
Week 2: Practise prompting
- Write prompts using role, task, context, constraints, and output format.
- Compare vague prompts with specific prompts.
- Ask follow-up questions to improve results.
- Practise summarisation, explanation, and structured output.
Week 3: Apply AI to your work or studies
Choose a practical area:
- Students: create revision questions and explanations.
- Developers: explain code and draft tests.
- Business professionals: summarise notes and draft reports.
- QA professionals: create and review test scenarios.
- Oracle HCM consultants: draft configuration checklists and test plans.
- Content creators: plan and edit articles.
Use non-sensitive examples and verify the results.
Week 4: Build a small project
Choose one repeated task and create a reusable AI-assisted workflow.
For example:
- A reusable prompt for meeting summaries.
- A test-case drafting template.
- A study quiz generator.
- A blog content planning workflow.
- A small coding project using an AI assistant.
Measure whether the workflow saves time and produces reliable output. Keep refining it based on actual results.
By the end of the month, aim to have one useful workflow you understand and can repeat—not simply a list of AI tools you have tried.
20. Frequently Asked Questions
Is AI only for technical people?
No. Many AI applications are designed for everyday users. You can begin with simple tasks such as summarising notes, asking questions, or drafting text without learning to code.
Do I need to pay to use AI?
Not necessarily. Many services provide free access with limits. A paid plan may be useful when you need higher limits, specific features, or professional controls.
Is Generative AI always accurate?
No. It can produce incorrect or outdated information, even when its response sounds confident. Verify important answers.
Is ChatGPT the same as AI?
No. ChatGPT is an AI application. Artificial Intelligence is the broader field that includes many different technologies and applications.
Is open-source AI always free?
Not necessarily. Software or models may have different licences, infrastructure costs, and restrictions. Review the licence and costs for your intended use.
Can AI replace software testers or consultants?
AI can automate or accelerate parts of their work, but requirements interpretation, professional judgement, validation, stakeholder communication, and accountability remain important. How roles change will depend on the work, tools, and organisation.
What is the difference between an AI assistant and an AI agent?
An assistant generally helps respond to requests. An agent may also plan and execute multistep tasks using tools and permissions. The distinction depends on the system’s actual capabilities; not every chatbot is an agent.
What should I learn first?
Start with AI fundamentals, practise clear prompting, learn how to verify outputs, and apply AI to one real task. Explore coding agents or automation tools when you have a specific need.
21. Final Thoughts: Learn, Experiment, and Grow with AI
Artificial Intelligence is a broad and evolving field. You do not need to master every concept or subscribe to every tool to benefit from it.
Start with the basics. Learn how different types of AI work, understand what the tools can and cannot do, and practise writing clear prompts. Then apply AI to tasks that matter to you.
Whether you are preparing for an examination, developing software, managing business activities, creating content, testing applications, or implementing Oracle HCM solutions, AI can become a useful partner in your work.
The most valuable skill is not simply knowing which AI tool to use. It is learning how to define a problem, provide the right context, evaluate the result, and take responsibility for the final outcome.
Use AI with curiosity, critical thinking, and care. Keep learning, keep experimenting, and share what you discover.
Learn. Grow. Inspire.
Thank you for reading Sadhana’s Journal. If this guide helped you understand AI, share it with someone who is just beginning their AI journey.