Pillar 2 · Put AI to Work

AI Agents & Enterprise Automation

How AI agents and automation actually work, where they help, and where humans must stay in the loop — with concrete examples, workflow diagrams, and honest limitations.

Chatbot vs AI agent — in plain English

Think of an ordinary chatbot as an assistant that answers your questions. An AI agent can also use approved tools to perform tasks — searching documents, updating a database, or preparing a report — and it still needs appropriate permissions, with important actions requiring human approval.

Diagram of an AI agent loop: a business request goes to the AI agent, which uses approved tools, verifies results, and escalates to a human for approval before the final action
The agent loop: tools are approved, results verified, and a human signs off on consequential actions.

A worked example: customer support triage

A company receives hundreds of customer support messages. Traditionally, a person reads and answers every one. An AI-assisted pipeline categorises each message, retrieves the relevant account information, drafts an answer, and escalates complicated or sensitive cases to a human. The AI drafts — a human approves anything consequential, and the system logs what happened.

Workflow diagram: incoming customer messages are categorised, account information is retrieved, a reply is drafted, and sensitive cases are escalated to a human reviewer
Triage workflow: automation handles the routine; people handle judgment calls.

What AI agents do well — and where they fail

Good at

  • Repetitive, well-defined tasks
  • Drafting and summarising at volume
  • Routing and categorising information

Needs humans for

  • Judgment calls and sensitive cases
  • Anything with legal, financial, or personal impact
  • Checking accuracy — agents can be confidently wrong

Permissions, privacy, error handling, and escalation paths matter more than the underlying model. Give an agent the least access it needs, log every action, and always keep a human approval step for consequential decisions.

Tools to start with

From the verified directory — plan details and limitations on each profile.

GitHub Copilot

AI Coding Assistants

An AI pair programmer inside your editor that suggests code as you type and explains unfamiliar code.

Zapier

AI Workflow Builders

A no-code automation platform connecting thousands of apps, now with AI steps inside workflows.

LangChain

AI Agents & Frameworks

An open-source framework for building LLM applications and AI agents — chains, tools, memory, and retrieval.