One-page guides

Each one is a single chart that explains an idea at a glance. Read it here, open it full size, or print it.

AI agents: from conversation to action

Tools, frameworks and concepts for building agents that do interactive tasks.

Chart titled “AI Agents: From Conversation to Action”. Panels cover what an AI agent is, five frameworks for building one, six agent capabilities, the loop from a human goal through context, reasoning, planning, tools and the environment back to feedback, the shift from traditional interfaces to goal-driven interaction, and a table comparing chatbots, assistants, agents and multi-agent systems. The whole chart is written out as text below this image.

Open it full size ↗ — the table is small on a phone, so the whole chart is written out below.

Read it as text

What is an AI agent?

A system that uses AI models, tools, memory and contextual information to pursue goals and perform interactive tasks.

Frameworks for building agents

  • LangChain — connects AI models to tools, APIs and data.
  • LangGraph — builds stateful agents with decision-making and workflows.
  • CrewAI — coordinates multiple specialised agents.
  • Microsoft AutoGen — supports multi-agent communication and collaboration.
  • OpenAI Agents SDK — builds agents with tool use, handoffs and tracing.

What an agent can do

  • Interpret a goal and plan the task
  • Hold context and memory
  • Use tools and APIs
  • Act on its own, or only once a person approves
  • Work with other agents
  • Take feedback, adapt, and recover from errors

The interactive loop

  1. A human goal
  2. Context and memory
  3. The AI model reasons
  4. Planning and decisions
  5. Tools and actions — APIs, apps, data
  6. The environment, real or digital
  7. Feedback — results, errors, new information

Feedback becomes updated context, and the loop runs again.

From interfaces to intent

Traditional interfaces are manual and step by step: click, type, select. Goal-driven interaction communicates an intention and the outcome wanted instead — “plan my trip”, “analyse this data”.

Chatbot, assistant, agent, multi-agent

ChatbotAI assistantAI agentMulti-agent system
Primary roleAnswer questions in conversationHelp with tasks and informationPlan, use tools and complete multi-step tasksSeveral agents collaborating on complex goals
AutonomyLowModerateHighDistributed and coordinated
Context and memoryLimited, short-termOngoing within a sessionPersistent and task-awareShared and specialised
Human oversightHigh, user-drivenHigh, the user guides tasksModerate: approve actions or set guardrailsModerate to high: set goals and monitor
ExampleA customer support botA personal assistant that schedules, drafts and searchesBooks travel, analyses data, updates a projectA research team of agents that plans, gathers, analyses and reports