“Think of agents as the new apps for an AI-powered world” - Jarred Spataro, Microsoft
An agent is able to independently perform tasks on your behalf.
They either do this according to pre-set flows and logic determined by you (simple agent), or autonomously to achieve predefined goals (AI agent).
Leaving aside much of the sound and fury around the term “agent”, the key things to know are:
- High utility: They’re very useful for automating internal and personal processes
- Time saving: A few simple (1-4 step) automations can save you a ton of time
- Time to value: They can be set up fast, giving you immediate value
- Limited technical skills required: There’s minimal prompting and coding involved
Another way to think about agents is that they are like apps with a backend, but no front end.
Agents process data behind the scenes, but take their inputs (e.g. an email is received, a spreadsheet updated, etc.) and make their outputs (e.g. send a Slack message, schedule a meeting, etc.) in existing apps (e.g. email, calendar, spreadsheet, etc.).
In this article we’re going to introduce the concept of agents, explain how they differ from bots and assistants, and walk you through how to create AI agents yourself.
We’ll also cover some general principles of when to use an AI agent and how to set one up. We’ll then walk through specific steps, including configuring triggers, selecting steps and testing and iterating.
Finally we’ll include exercise steps for you to complete, with step-by-step video walkthroughs. You’ll build simple calendar based automations, and start using AI to power these.
What is an AI agent
We’ve become more and more familiar with interacting with LLMs via a number of interfaces:
- Chatbots: Conversational AI interfaces. E.g. ChatGPT, Claude, Grok.
- Assistants: Embedded AI in existing tooling that helps with tasks. Think Copilot, interacting with Gemini in Google Workspace or Search, and Amazon’s Alexa.
- Agents: AI which is embedded in workflows to carry out tasks. Most of these are custom internal tools that follow the pattern when [trigger] do [series of actions]. Common tooling to build agents includes n8n, Zapier and Relay.
These are all interfaces, or ways of interacting with a relatively small number of LLMs, usually a model from OpenAI, Anthropic, Google or Meta.
They’re all now available both as products from the companies which own the LLMs, and as potential products for you to serve to your internal and external customers, or to enhance your own productivity.
Definition of Agents
Agents differ from chatbots and assistants in a couple of key ways.
The major one is that rather than instigating a conversation or task, you can set them up to run in the background repeatedly, and without your constant engagement.
One of the biggest advantages of any agent type, be they simple agents or workflows, or AI agents, is that they will work for you while you’re off doing something else.
They do this by combining LLM and rules based reasoning to actually take complex actions in the real world, through integrations, tools, and APIs.
AI agents differ from earlier automations like Zapier - which performed tasks in the background according to preset logic - in two major ways:
- Can perform against a much wider set of use cases: by using LLMs as a node in workflows, AI agents can handle many more situations, react in more personalised ways, and output generated text, images, video and audio.
- Can build decision trees faster: AI support means that setting up automations becomes more user-friendly and less time-consuming.
This distinguishes them from existing types of workflow automations which operate according to preset logic to get to a result.
Another way to think about this is:
- Chatbots: Respond when asked
- Assistants: Monitor, suggest, respond when asked
- Workflows / simple agents: Work in the background following rules - if this, then…
- AI agents / complex agents: Work independently and make contextual decisions based on goals
All of them are great for different purposes. The key thing is to understand which you need for which task.
In reality there’s lots of overlap but understanding the distinctions helps to choose the right tool and knowing what tool is best for what is key when you're thinking about how to create AI agents. Knowing which to use and when can help you avoid over-engineering the things you want to automate.
