Skip to main content
Technology

From HubSpot Workflows to AI Agents: When Should You Use Which?

The difference between HubSpot Workflows and AI agents: rules versus judgement. Five sales tasks, when to use which, and how a person approves the result.

Author: Agent360 editorial team
Published: 21 April 2026
Last updated: 7 November 2026
Reading time: 10 min read

How do AI agents differ from HubSpot Workflows?

A workflow in HubSpot follows rules: "If a contact fills in the form, send email X and create task Y." It does exactly what it is set up to do, every time. An AI agent can instead read free text, interpret a situation and write a draft or an assessment. In return, it does not always give the same answer, and it can be wrong. They solve different kinds of tasks, and the best solutions use both.

TL;DR:

  • Workflows are predictable, cheap and easy to control. They are the right choice when the rules are clear.
  • AI agents are useful when the input is free text, or when a task requires judgement. They need a control: approval and a log.
  • Combine them: the workflow directs the process, an AI step reads or writes, and a person approves where it matters.

What are HubSpot Workflows, and what can they not do?

Workflows are HubSpot's tool for automating processes based on triggers and conditions: when something happens, the system does something. Typical examples are assigning leads to the right salesperson, sending an email on a particular action, creating tasks, or updating a property when a deal changes stage. Which functions you have depends on your subscription.

The strengths are clear:

  • Predictable. The same input gives the same result.
  • Transparent. You can read what happens and why.
  • Cheap to run.

The limitations are just as clear:

  • They cannot read free text and understand what the customer means.
  • They are brittle when faced with exceptions. Each exception requires a new rule.
  • They do not know what was said in a meeting, unless someone has typed it in.

What are AI agents, and what makes them different?

By an AI agent we mean here a solution in which a language model is given a task, something to read, and some rules for what it may do. It can, for example, read an email, understand what is being asked, and write a draft reply. Or read meeting minutes and suggest which CRM fields should be updated.

This gives new possibilities, but also new risks:

  • It can be wrong. Especially with numbers, names and agreements.
  • Answers vary. The same input does not always give the same output.
  • It costs per run. The amounts are small each time, but they add up.
  • It must be checked. You must be able to see what it has done, and to step in.

For that reason, an AI agent in sales should normally suggest, not decide.

Comparison: Workflows versus AI agents

Question HubSpot Workflow AI agent
What does it build on? Rules and conditions Interpretation of text and context
Is the answer always the same? Yes No, it can vary
Can it read free text? No Yes
How easy is it to control? Easy Requires approval and log
What is the risk? That the rules are wrong or incomplete That it interprets wrongly
What does it cost to run? Very little Little per run, depends on use
What is it good at? Fixed processes Text, judgement and drafts

When should you use workflows, and when AI?

A rule of thumb:

  • Can you write the rule in one sentence without "it depends"? Use a workflow.
  • Does the input come as free text, or does the answer require judgement? Consider an AI agent.
  • Are the consequences of an error large? Let a person approve, regardless.

Many believe that AI must replace workflows. That is seldom right. A workflow is still the best way to send an email on a particular event. AI is better at reading the email the customer replies with.

Five examples from sales

Example 1: Updating the CRM after a meeting

Workflow alone: Can create a task for the salesperson to update the CRM. That moves the problem, but does not solve it. With AI: An agent writes the minutes and proposes fields and next steps. The salesperson approves, and only then is it written into HubSpot. Do this only if the CRM is already used consistently. The human: Approves, and everything that goes to the customer.

Example 2: Follow-up after a meeting

Workflow alone: Sends a standard email after x days. With AI: Writes a draft that builds on what the customer actually said. The human: Reads it through and sends it themselves.

Example 3: Inbound leads with free text

Workflow alone: Distributes by fields such as country, size and source. With AI: Reads the message in the contact form, understands whether it is a sales enquiry, a support case or spam, and proposes a category. The human: Checks the uncertain ones.

Example 4: Deals that appear to stall

Workflow alone: Marks deals with no activity for x days. With AI: Can read recent emails and notes and suggest why the deal has stalled, and what the next step could be. It is a suggestion, not a prediction. The human: Decides what should be done.

Example 5: Preparing for a customer meeting

Workflow alone: Sends a reminder. With AI: Gathers earlier correspondence, notes and public information into an overview. Anything it finds outside the CRM must have a source, so the salesperson can check it. The human: Reads, assesses and adapts.

How to combine them

The pattern we recommend has four parts:

  1. The workflow directs the process. It determines when something happens.
  2. An AI step does what requires text understanding. It should return a structured result (for example a category and a reason), not just free text, so it can be used further.
  3. A person approves where it matters. Price, stage, agreements and anything that goes to a customer.
  4. Everything is logged. What came in, what went out, and who approved.

HubSpot can call other systems from a workflow, for example via webhooks and code actions, depending on your subscription. That is often how an external AI solution is connected.

How to design an approval step

An approval step is what makes it possible to use AI without fearing it. It must be easy to use, otherwise it will be skipped.

  • Show the proposal and its source together. The person approving must be able to see what the AI proposes and what it has based it on, without searching.
  • Make "approve" one click, and "correct" just as easy. If correcting is cumbersome, people approve instead of reading.
  • Mark the uncertain. When the AI is in doubt, or when information is missing, this must be stated clearly.
  • Let high-risk fields always require a yes. Price, agreement stage, probability and anything that goes to a customer.
  • Save what was changed. Corrections are the best source for improving the instructions.

How to decide whether the AI step is good enough

Before an AI step is allowed to run in daily use, it should be tested on your own examples.

  1. Collect 30 to 50 real examples of the task, for example emails, forms or meeting notes.
  2. Have a person write down the right answer for each example.
  3. Run the AI step and compare. Count how often it is right, partly right and wrong.
  4. Look at the wrong ones: are they harmless, or do they cost something? The more expensive an error, the more a person must check.
  5. Set a limit, and repeat the test when you change instructions or switch model.

We have written a longer guide to such a test in the guide to testing a solution on Danish sales language.

What can Agent360 contribute?

We build AI solutions in the customer's own systems, including on top of HubSpot. We contact your suppliers and help with access, and you are copied on everything throughout. We build one flow at a time and finish the first completely before we go further. You decide what the system may do by itself. Everything else waits for an approval.

On the use cases page you can see the example "The meeting is over. The CRM work begins", where the agent may write the minutes and update the fields, but asks first before anything goes out to the customer. It can be connected to HubSpot, Salesforce, Pipedrive and Microsoft Dynamics, as well as telephony and calendars. The examples use invented names and figures and show the structure, not a result.

How do you calculate whether it is worth it?

There are no universal figures. Calculate on your own:

  1. How many times a week does the task occur?
  2. How long does it take today, and who does it?
  3. What does that time cost? Multiply by an hourly wage.
  4. What does the solution cost to build and run? Include maintenance and time for approval.
  5. What is the error rate, and what does an error cost?

Compare afterwards with what you measured before. The price of what Agent360 builds depends on the task. We give an estimate after a short conversation.

FAQ: HubSpot Workflows and AI agents

Q: Do AI agents replace workflows? No. They complement one another. Workflows are better at fixed rules, and AI agents are better at text and judgement. The best solutions use both.

Q: Can HubSpot run AI agents itself? HubSpot has its own AI, Breeze, with functions that vary with subscription. See our guide to HubSpot and AI. If your task is special, a custom solution may supplement it.

Q: Is it safe to let AI write in the CRM? If you limit rights, let it write notes before fields, require approval for stage, price and probability, and log everything, it can be. Without those things it is not safe.

Q: What happens if the AI gets it wrong? That is why there is approval and a log. If something could have major consequences, a person should always say yes first. Errors are corrected in rules and instructions, and you can go back through the log.

Q: How long does it take to set up? It depends on the task and on how tidy your data is. Start with one task and finish it.

Q: What does it cost? It depends on the task: how many systems are connected, how much approval logic has to go in, and how much data has to be cleaned up. We give an estimate after a short conversation.

Conclusion: Workflows are the foundation, AI is a supplement

Use workflows for what can be written as rules, and AI for what requires text understanding. Let a person approve where it matters, and log it all. Start with one task, and build it finished.

Book 20 minutes of free advice if you want to hear which of your tasks is worth building. We will also tell you if none of them are. See also examples of tasks we can help with.

#hubspot#workflows#ai-agents#automation#sales

Should your CRM keep itself up to date?

We build the layer that moves data into the CRM you already use, without anyone having to type it in. We look at your current set-up first.

How we built it · CRM & customer overview

Customer CRM that updates itself

An internal page per customer with contacts, the agreement, the projects and a timeline where emails, meetings, email drafts and tasks land by themselves. Above the pages sits the board with all customers, owner and latest contact, and the system suggests next steps based on fixed rules.

  1. Email, meeting, task
  2. Lands on the customer’s page
  3. You take the next step
See the customer’s own page →