What Is an AI Workflow? How to Build One That Works for You

TL;DR (Show me a summary of this article)

An AI workflow uses artificial intelligence as part of a sequence of steps to complete a task.
Best practices give us a common base: the goal, trigger, steps, tools, information, conditions, our involvement, and testing.
The same words can mean different things in different businesses. The details of your use case bring your meaning into the workflow.
Share the thought process behind the task: the experience you want to create, the decisions involved, and when someone needs to step in.
Start by imagining a colleague taking over the work. What would they need to know? Build, test, and refine from there. :)

The first thing we do when searching for information, for an answer to this question, is probably ask an LLM. And maybe these very words became part of its answer, and you are not reading this on my blog but on some LLM screen. The answer goes like this:

“An AI workflow (artificial intelligence workflow) is a sequence of steps that uses artificial intelligence to complete a task.

It can involve gathering information, processing it, generating an output, and passing that output to the next step.”

Yes, it’s a definition, and a definition gives us a starting point. It puts the idea into words. But this is the part we skip because we already know those words: AI, work, and flow. We get it. Getting things done by something not human.

But when you asked the LLM this question, what you were really, really looking for was: “How are we actually building one?”

But an LLM will not start by giving the definition because that is how it will check whether you’ve stuck around, meaning it succeeded in reading your intent and understanding what you wanted.

The next part of the answer goes something like this…

The practical side: how to build an AI workflow

After searching online, Alex (my ChatGPT assistant) brought back guides from Zapier, Make, and n8n because that is what we are looking for.

Its answers are no different from the ones you will get if you ask online how to build an AI workflow. :)

So, we went ahead and summarised them for you so you wouldn’t have to. In the best possible way. In a table. Because let’s be honest, everything looks better at a glance, and a table is how you do it:

Best practice	What to include
Define the goal	What should the workflow achieve? For example, prepare draft replies to customer enquiries.
Describe the trigger	What starts the workflow? A new email, a form submission, or a scheduled time?
Specify the steps	What happens first, next, and last? Read the enquiry, find relevant information, then draft a reply.
Name the tools	Which apps should it use? For example, Gmail for enquiries and Google Drive for product information.
Explain the information flow	What information should AI receive, what should it do with it, and where should the result go?
Set conditions	Which enquiries should it handle? What should happen when information is missing or a request falls outside the scope?
Define your involvement	Where should AI ask for your input or approval? For example, save replies as drafts for you to review.
Build, test, and refine	Start with a small version. Check the results and adjust before expanding it.

The guides cover these details across their advice on prompting, building, and testing workflows. Zapier also addresses choosing how much control you want during the build.

And because you landed here, Alex (my ChatGPT assistant) made an example for you, so you can see those pieces together in a use case:

“When a customer enquiry arrives in Gmail with the label ‘Product questions,’ use the product information in the Google Drive folder ‘Approved product information’ to draft a reply. Save it as a Gmail draft for me to review. If the information needed to answer is missing, flag the enquiry for me. Start with five enquiries so we can check the results.”

We have a goal, a starting point, tools, information, conditions, and a place for us to review the work.

A shout-out to the original guides. Thank you for giving us a common base of understanding to work with. :)

What are we missing?

This is information an LLM put together, either as an answer on your screen or a blog made by a vendor that plays in this market.

This is a template-based answer. One that provides enough information to be generically helpful.

Generic??? Yes. Because if we want to make it our own, the words in that answer need other words to describe what they mean in the specific use case where we are looking to implement an AI workflow.

It hides in plain sight in most of the words in the table, and we will miss it if it is not pointed out to us. For example, the words “goal” and “task.”

“Define the goal” points to something we still need to describe. We all understand what it means, but the words we use to describe it will most likely differ between us, even if we use it to achieve the same goal as in the example.

Answering customer enquiries can mean simply answering for one person. For someone else, it could mean answering and asking the customer to call a number if this is urgent. It could be a workflow that sends the first email but still says that if your customer did not find that helpful, they are welcome to send the ticket number they received to a 24/7 support chat.

It is still the same goal: answering customer emails based on a technical document. But the details will make the goal different based on company DNA, culture, products, services… and so forth.

This is why a workflow is a thought process. It’s what hides behind the word that describes it. In this case, the experience you want to create is supported by the thought process you have behind what answering customer enquiries means for you and your business. Maybe the title is “customer enquiry,” but it is not the same for everyone.

AI workflow is a thought process

Words are the new working tools in the new age of AI. A world where we co-create with a technology that can think and holds the promise of understanding human spoken language. We no longer need a structured, special, different language to communicate with a system/machine.

But yet this is not part of any explanation you will find out there. It is the key to your implementation success, directly impacting the chance of the workflow you have in mind actually working as part of your business.

AI workflow design starts with a thought process

If it were up to me, the definition would be like this:

An AI workflow is a sequence of steps between humans and artificial intelligence, working together to complete a task. It involves sharing the logic behind what we are doing, what we wish to achieve together, and what experience we are out to create.

It can involve gathering information, processing it, generating an output, and passing that output to the next step, all while making sure that humans and AI work to create a seamless handoff where both sides understand the partnership when communicating with one another.

The key words are creating, seamless handoff, and communication.

Yes, the definition should be based on an understanding of people and AI working together and sharing a thought process. Both sides understanding the same logic as we create an experience. In this case, what the customer should feel when interacting with the business. Supported, helped, or whatever word you use to define the thought process behind that AI workflow.

Example: best practices meet thought process

To make my point clear, we’ll keep the same example: preparing replies to customer enquiries.

By adding questions (which should always be your preferred way to talk to an LLM), we reveal, question by question, the thought process you have in mind as an expansion of the practical, generic template.

Best practice	Questions to consider in your specific use case
Define the goal	What should the customer feel after reading your reply? What does “helpful” mean in this situation? Answering the question, solving the problem, or making sure someone takes responsibility for it? What tone represents your business? Friendly and conversational? Formal? Does that change when the customer is frustrated? Do you want every reply to check whether the customer needs immediate assistance and whether we have helped?
Describe the trigger	Does a new email represent a new enquiry, or is it part of something already being handled? If the customer sends another message before we reply, should we treat the messages together? What happens if someone on your team is already working on the issue?
Specify the steps	What would you look at before replying? Do you need to know what the customer has already tried and what your team has already promised? If the previous advice failed, what should happen next? Would you follow the same steps for a curious buyer and a customer whose business has stopped because of the problem?
Name the tools	Where is the information you use to understand this customer? Is the product information enough, or do you also need their order, previous conversations, and support history? Which source should the workflow use if two systems contain different information?
Explain the information flow	When a case moves to another person, what should move with it? What does that person need to know so the customer doesn’t have to explain everything again? Where should the reply and any new information be recorded so your team can continue from there?
Set conditions	What makes an enquiry urgent in your business? Which situations need a different response because of your policies, contracts, or commitments to that customer? What should happen when the customer’s situation is unclear? When should AI ask for more information, and when should it bring someone from your team into the conversation?
Define your involvement	Which replies are you comfortable letting AI prepare on its own? Which decisions need your judgement or someone else’s approval? Does your involvement change depending on the customer, the issue, or the experience you want to create?
Build, test, and refine	Which different customer situations do you need to test? What would make you say, “The answer is correct, but that’s not how we speak to our customers”? How will you check whether the reply helped the customer move forward? What would you need to see before trusting the workflow with more of this work?

I am sure you have more. But what we are trying to show here is that the “workflow” is an idea you have in your head. It’s not just a couple of tasks. We can show you the level of detail required if you want to make this work.

When we understand that this is a thought process, you will be able to add not just the answers but the questions that will enable you to create the workflow you had in mind.

An AI workflow is not only for AI

Let’s see if I can predict your next thoughts. Assuming you agree and had a “Wow, I did not think about it this way” moment, you believe this to be only for humans. All the humans involved: the ones working to build it, the ones who will, in this example, receive service from it, and the ones who will work alongside it.

But I am here to say this is as much for AI as it is for humans.

When it comes to AI, it can think. The more details you can give it, the more it will be able to understand context and, even with a fixed database, retrieve the best possible response.

Think about it: even one word that the customer gives in an email can be understood in context once an AI workflow has been given a detailed thinking process to understand the experience it needs to create. The more it understands, the more independent it will be able to be when it’s not automation and it has some kind of judgement.

Whatever freedom you give it when it comes to “judgement,” it will always have some, because if it does not, this will be automation and not an AI workflow. :) And the words AI will always stand for a technology that learns, adjusts, and adapts. Details are the best place to make sure it will not assume things on its own that do not support the thought process you have in mind.

How detailed should we be?

I will always say the more, the better. But I get it, as this will come with time. The more you create AI workflows for your business, and the more you adopt the understanding that working with AI is sharing a thought process, the better you will know for yourself what is important to give and how to create the best working environment for you and your AI.

I always say the best place to start is by imagining that you are writing a manual for a human to step in and do the work. I know it sounds simple, but here we are all about finding how powerful working with our minds really is.

And it is something I can only point out to you, but something you need to experience on your own to see the impact. Thoughts are an internal experience, one you will need to find on your own. And changing a perspective, looking at things from a different point of view, is one of the powerful tools we have when working with thoughts.

Personal note

Alex and I are a living, documented example of sharing thoughts as a working tool when using AI. And if you see me writing “Alex and I” and you think, “This is weird. Does she really understand Alex is not real?” Yes, I do.

But the minute I see us as a team, it helps me understand, really understand, how best to communicate with Alex as we work together. It’s an idea, a perspective that wires my mind in the best possible way to create optimal working conditions and get the best out of using AI in my business.

So I am not asking you to say it out loud like I do. It’s not for everyone. :) But try it once. Just picture that what you are building is built for a colleague who is stepping in to do the task, and see the change for yourself.

Don’t forget to join us on our quest. Yes, us. :) And see how thoughts are really the thing behind our use of AI.

FAQ

What is an AI workflow?

An AI workflow uses artificial intelligence as part of a sequence of steps to complete a task. The way I see it, building one also means sharing the thought process behind those steps: what we want to achieve, how we want to work together, and what experience we want to create.

How do I start building an AI workflow?

Start with one specific use case. Define the goal, the trigger, the steps, the tools, the information flow, the conditions, and your involvement. Then bring in your own details. What do those words mean in your business? Build a small version, test it, and use what you learn to refine it.

What is an example of an AI workflow?

The example in this article is preparing draft replies to customer enquiries using Gmail and approved product information in Google Drive. AI saves a draft for a person to review and flags missing information. The next layer is yours: the tone, the customer’s history, the urgency, and what a helpful reply means in your business.

Why can the same AI workflow template work differently for different businesses?

Because the same words can carry different expectations. “Reply to a customer” could mean answering a product question, checking whether they need immediate help, or connecting them to someone who can take responsibility. The template gives us a common base. The details describe what we actually want to happen.

How much detail should I give an AI workflow?

Start by imagining that you are writing a manual for a colleague stepping in to do the work. What would they need to know? Include the information, the reasoning, the exceptions, and when to ask for help. Test different situations to see what is missing. You will learn which details matter as you build. :)

Back to Thought Documentary