Something my brother Phil and I would do before I had kids was travel to go to concerts. We’d go with a couple of friends, grab a hotel room, and enjoy the show.

One time after check-in and before the show, we decided to grab some food. And since we were college students with limited funds, we got a box of Pop Tarts. And since the hotel didn’t have a toaster, I used the microwave.

I’m not sure if you know this, but the instructions say to microwave a Pop Tart for 3 seconds. I set it for 2 minutes. That was a mistake.

As you can imagine, the Pop Tart was burned to a crisp and the smoke alarm went off. All because instead of reading the instructions or using a more appropriate tool, I used something too powerful.

I’m noticing this trend with automation and AI; people are starting to use the terms interchangeably. But they are different, and using AI for a simple process automation is kind of like microwaving a Pop Tart.

What’s The Difference Between Automation vs. AI?

From a purely “common use case” standpoint, automations make things happen based on some inciting incident, called a trigger. An example I use commonly is, If it’s sunset in December, turn on the Christmas lights. Sunset is the trigger. Turning on the Christmas lights is the action. There’s a “condition” to only do it in December. This is very predictable. You can look at the month and know if and when the Christmas lights will turn on.

AI has largely been used as a program you can “chat” with to find information, process data, or do other things you probably shouldn’t use it for.

What makes this confusing is that LLMs can be used for automation, and are increasingly being used as such thanks to tools like Claude Cowork. So to really differentiate between the two, I want to borrow a term I learned from Emily Mabie over on LinkedIn.

Traditional, pre-ChatGPT era automations (think Zapier, Make.com) are Deterministic Automation. You know the input and the output, and the process runs exactly the same every time.

I’m referring to automation via LLMs/AI (or AI “agents”) as Interpretive Automation¹. The scope of the task is broader, and the automation is run by an interpretive algorithm, which means that the output may change slightly (or dramatically) each time.

So the difference between Automation vs. AI in this context is that Automation is “Deterministic.” AI is “Interpretive.”

When You Should Use Deterministic vs. Interpretive Automation

Now that you know the difference, when should you use one over the other? Here’s a simple test.

The 4 components of deterministic automation are triggers, actions, conditions, and timing. Think about these when you’re considering which to use. Then ask yourself these questions:

1. Does this always have the same input (trigger)?
2. Does it always have the same result (actions)?
3. Is there little to no decision-making when it comes to the input and output (conditions)?
4. Do I need it to run consistently (timing)?

If the answers to all of these are yes, you need deterministic automation. On the other hand, if the input, output, or decision-making are not straight-forward, interpretive automation is the way to go.

In my opinion, most business automations would be better served by deterministic automation.

Why Deterministic Automations Are Better for Solopreneurs (and Most Businesses)

Automations are best when they are predictable and reliable. By definition, interpretive automations are not that, as the AI might change how the algorithm interprets an instruction.

With deterministic automation, when the trigger happens, the actions happen every time — and predictably. When a deterministic automation tool does fail, it does so “loudly”…that is, it will tell you when it errors because it’s very clear on what an error is. You’ll usually get an email, and then if it fails enough, the automation will turn off and you’ll get an email about that too.

Now, something that AI can do is sometimes verify the results by “looking” at them. This is helpful, but also not consistent (and unnecessary if you know what the results will be).

A couple of other drawbacks of interpretive automation are that errors may not be surfaced at all…especially errors with the results. AI is sycophantic, which means that it will agree with you for each individual pushback. That feedback may not impact the next run though.

Those probably can be fixed with better instructions (“fail loudly,” “don’t just agree with me, verify the end state”). But there’s one drawback that none of us can do anything about.

Interpretive Automations Can Change with the Model

When the LLM model changes — and therefore the engine running the automation — the automation can also change. We’ve seen this a bunch of times over the past year especially. We even have leadership at Claude recommending resetting your LLM’s memory with every model upgrade.

I experienced this with a few skills in Claude, where the model changed and the results of the skills changed…which means I had to change the skill.

If you’re doing something your business relies on, you should ask yourself: “if this breaks suddenly, how much time am I willing to invest in fixing it?”

Conversely, deterministic automation tools are very reliably backwards compatible. When something changes, those changes are communicated well ahead of time, and when the change is made, things don’t break immediately.

That’s the kind of reliability you should have with your business automations.

Examples of Deterministic Automation vs. Interpretive Automation

I’ve already mentioned the Christmas lights example as a good deterministic automation. What about some related to your solo business? Here are a few that are perfect for deterministic automation:

• When I add a new contact in Notion (trigger), send a new task to Todoist (action)
• When someone books a clarity call with me (trigger), tag them in Kit.com (action)
• When I update a status in Notion (trigger), send an email to my VA (action)

All of these are straightforward, predictable, and happen the same every time.

Now a couple of interpretive automations:

Scan my email inbox for tasks. This one is interpretive because there’s no clear trigger; I get emails throughout the day. There’s also no obvious way to find tasks. It takes the interpretation of an LLM to determine what is or is not a task.

Review and score emails I sent to my mailing list over the past week. This pulls emails over a rolling timeframe and applies a score based on several factors (opens, clicks, unsubscribes). Could I do this with deterministic automation? Possibly…but it would heavily depend on what Kit.com (my ESP) exposes to Zapier.

Create assets and upload any new podcast episodes to RSS.com. This one looks at the status of episodes in Notion. If they’re marked “Ready for Publish,” the LLM grabs the transcript, creates a basic description and gathers show notes, then uploads everything to RSS.com.

This one technically could be done with deterministic automation. In fact, there’s a very clear trigger (status change), condition (status is “Ready for Publish”) and end state (episode is uploaded). However, there are a lot of factors in making it run smoothly:

• I need to essentially write code to work with the RSS.com API.
• I need to host the videos somewhere Zapier can access and then upload.
• I still need to call some LLM to process the transcript.
• I need to give a lot of context to the LLM and/or Zapier.

In this instance, creating an interpretive automation on my computer that can access all the files was better. It also used Claude Code to write a script to talk to the RSS.com API. That script will not change, so the main mechanism for the end state is still intact, even if models change.

It’s All About Conscientious AI Use

The main takeaway here isn’t that you should only use one or the other. It’s that you should be conscientious about which you choose. Are you just blindly giving things to an LLM because it’s easier? Because you haven’t thought about it?

Or because you don’t know much about deterministic automation tools like Zapier?

Interpretive Automation and AI can be great for a lot of things, but you need to do what will be most reliable for your business…especially if you’re a solo business owner.

Footnotes