Learn AI

Learn AI covers the ideas behind the tools and follows them into connections, workflows, and responsibility. Start Here gives you a concrete first step. Then pick the group closest to your question; you don’t need to read the section in order. Use the explanations, examples, and short exercises to connect the vocabulary with practice. Follow related pages for more detail. For your next brief, the Prompt Lab can help turn that understanding into a useful task.

Start Here

AI starts with a question, not a tool.

A clear starting point for anyone who wants to understand AI without memorizing a wall of acronyms.

Explore AI starts with a question, not a tool.

Foundations

For readers who want clear meanings for the terms used in AI conversations.

Models, assistants, and products

Understand the engine, the application around it, and the features that turn a model into something useful.

Explore Models, assistants, and products

The AI landscape: a map, not a ladder.

Methods, capabilities, and systems overlap. Knowing the distinction makes the headlines easier to read.

Explore The AI landscape: a map, not a ladder.

Generative AI is good at possibilities. You provide the standard.

Understand generated text, images, audio, and code without confusing fluency with reliability.

Explore Generative AI is good at possibilities. You provide the standard.

Training, inference, and fine-tuning

Separate learning a model from using it, and understand when extra training may be appropriate.

Explore Training, inference, and fine-tuning

Local, cloud, and open models

Choose where a model runs separately from who provides it and what its license permits.

Explore Local, cloud, and open models

Prompts, context, and tokens

A good brief gives the system a job, useful information, boundaries, and a way to recognize a useful answer.

Explore Prompts, context, and tokens

Why fluent answers can be wrong

Treat a convincing answer as something to examine, especially when it includes facts, quotes, numbers, or decisions.

Explore Why fluent answers can be wrong

Reasoning, capability claims, and AI limits

Judge what a system demonstrates on a task without treating a product label as evidence of understanding or general intelligence.

Explore Reasoning, capability claims, and AI limits

Structured outputs and validation

Turn an answer into fields software can use while checking meaning, uncertainty, and missing information.

Explore Structured outputs and validation

What AI can work with

For anyone working with data, pictures, speech, or systems that act in the physical world.

Predictive AI and structured data

AI is also classification, forecasting, recommendations, and decisions informed by patterns in data.

Explore Predictive AI and structured data

Computer vision: understanding visual input

Explore image classification, detection, extraction, and interpretation without assuming a system sees as a person does.

Explore Computer vision: understanding visual input

Speech, voice, and audio AI

Distinguish transcription, speech generation, translation, and a voice agent that can act.

Explore Speech, voice, and audio AI

AI can work across text, sight, and sound.

What multimodal means, and a useful way to test the limits for yourself.

Explore AI can work across text, sight, and sound.

Robotics and physical AI

When AI can move through the world, errors affect equipment, places, and people—not just a text response.

Explore Robotics and physical AI

Knowledge and connections

For readers who want to understand how an assistant finds information and reaches other software.

Embeddings and vector search

Understand how a system can find related material even when a question uses different wording.

Explore Embeddings and vector search

Give the model a source. Still check the answer.

Retrieval-augmented generation connects responses to reference material. It does not make every answer true.

Explore Give the model a source. Still check the answer.

Knowledge quality and access boundaries

Good answers need current, relevant information that the requesting person is allowed to see.

Explore Knowledge quality and access boundaries

Plugins, connectors, and skills

Separate packaged capabilities, reusable instructions, and the connections that let AI reach other software.

Explore Plugins, connectors, and skills

MCP: a common way to connect AI

Understand hosts, clients, servers, tools, resources, and prompts in the Model Context Protocol.

Explore MCP: a common way to connect AI

APIs, webhooks, and app connections

Understand a request to another system, an event that starts a process, and the credentials that make it possible.

Explore APIs, webhooks, and app connections

Agents and workflows

For people planning a process in which AI can choose steps or use tools.

An agent needs boundaries before it needs more tools.

What changes when AI can act, and how to introduce it without handing over the whole workflow.

Explore An agent needs boundaries before it needs more tools.

Inside an AI agent

The model is one part. The job, tools, state, checks, and stopping rules make it a working system.

Explore Inside an AI agent

Tools and function calling

A model can request an action. The application decides how to validate, authorize, and execute it.

Explore Tools and function calling

Agent memory and state

Understand what is remembered, where it is stored, and how it changes the next decision.

Explore Agent memory and state

When more than one agent helps

Specialized roles can divide work, but coordination adds cost, complexity, and new failure modes.

Explore When more than one agent helps

Sandboxes and controlled environments

Give a system space to work while limiting the files, network access, and authority available to it.

Explore Sandboxes and controlled environments

Designing an AI workflow

Make a useful task repeatable with clear inputs, steps, outputs, exceptions, and human checkpoints.

Explore Designing an AI workflow

Common workflow patterns

Recognize chaining, routing, parallel work, feedback loops, and orchestration without making a process unnecessarily complicated.

Explore Common workflow patterns

Safety and control

For anyone deciding what an AI system may do and how people can review or stop it.

Permissions and meaningful approvals

Decide what a system may read, what it may change, and when someone needs to authorize a concrete action.

Explore Permissions and meaningful approvals

The human in the loop needs a real job.

A review step only helps if the reviewer has the context, time, and authority to change the outcome.

Explore The human in the loop needs a real job.

Prompt injection and untrusted content

A document, page, or tool result can contain instructions that try to redirect the system away from its real task.

Explore Prompt injection and untrusted content

Privacy across the AI data lifecycle

Follow information from collection and prompts through connections, logs, memory, retention, and deletion.

Explore Privacy across the AI data lifecycle

Evaluations: does the system do the job?

Create representative tests for quality, tool behavior, refusal, escalation, and the effects of a change.

Explore Evaluations: does the system do the job?

Monitoring, costs, and incident response

A system needs an operating owner, visibility into behavior, resource limits, and a way to recover.

Explore Monitoring, costs, and incident response

Governance and society

For readers considering responsibilities, access, public trust, and the wider effects of AI.

AI governance: who owns the outcome?

Turn good intentions into named responsibility, clear boundaries, evidence, and decisions throughout the life of a system.

Explore AI governance: who owns the outcome?

An AI use policy people can follow

Write practical rules for tools, data, review, disclosure, and getting help without turning policy into unexplained jargon.

Explore An AI use policy people can follow

Reviewing an AI vendor or extension

Assess a product against your use case, information, account settings, operating needs, and ability to leave.

Explore Reviewing an AI vendor or extension

Fairness, bias, and decisions about people

Look beyond an average score to who receives the benefit, who experiences mistakes, and who can challenge a result.

Explore Fairness, bias, and decisions about people

Accessibility and human agency

AI can open doors when people can understand, control, correct, and decline the experience.

Explore Accessibility and human agency

Synthetic media, consent, and provenance

Explore generated images, video, and voices while keeping the audience’s understanding and the subject’s permission in view.

Explore Synthetic media, consent, and provenance

AI, energy, and environmental tradeoffs

Look at computation, infrastructure, and the useful outcome when judging the environmental cost of an AI system.

Explore AI, energy, and environmental tradeoffs

Brain–computer interfaces: communication, not instant expertise.

A real research signal—and the difference between clinical assistance and imagined human enhancement.

Explore Brain–computer interfaces: communication, not instant expertise.

Share this page