Automating AI workflows can dramatically increase your productivity, but it comes with risks. What happens when AI generates an inappropriate response or makes a mistake? You don't want to automatically send incorrect or awkward messages to clients and colleagues.
Fortunately, you can add human oversight to your AI automations without investing in expensive enterprise software.
In this guide, we'll explore two approaches to creating human-in-the-loop automations: Make's native enterprise feature and a free alternative using Airtable. Let's dive in.
Option 1: Make's native “Human in the Loop” feature
Make (formerly Integromat) recently introduced a dedicated human-in-the-loop module for enterprise subscribers. Here's how it works in a simple setup with two automations involving an AI prompt.
1. Trigger your first automation
Launch your first automation however you’d like. For instance, the scenario pictured below triggers when there’s a new row in a specified Google Sheet.
You can also add any other actions or searches you want to include to populate your
2. Generate AI content
Send a prompt to AI tools like OpenAI or Claude to generate a response. Be sure to include any necessary data from your trigger or other previous modules in your prompt.
3. Add a “Human in the Loop” module
Add a “Human in the Loop” module, and choose “Create a review request” as the action.
With this module, you can create a simple webpage hosted by Make that will include your AI response (as well as any other content you want to review).
Just add items in the “Content to review” section to add them to the page. In our example pictured below, we’ll add the AI response as an item to review.
Be sure to mark the content as “editable” if you want to be able to make changes to it on the review page.
Under “Advanced settings”, you’ll see several options for styling the page.
Once you’ve configured the module to your liking, click “Save”.
NOTE: while you can add this module to your Make scenarios on any plan, you’ll need an Enterprise plan to actually run it.
4. Share the “Human in the Loop” URL
Add another module to share a link to the Human in the Loop review page wherever you’d like. For example, in the automation pictured below, the link is sent to a Slack channel.
5. Run the automation and approve, deny, or modify the content
When this first automation runs, it will create a page where you can review the AI-generated content, and send the link to your desired destination.
Open up the link, review the content, and approve, deny, or edit the content as appropriate.
6. Trigger additional automations on approval/denial
Create a second automated scenario. For the trigger, choose “Human in the Loop”, and select “Watch completed reviews”.
This trigger will launch whenever someone takes an action on a “Human in the Loop” request (e.g., Approve, Deny).
While powerful, this feature is limited to enterprise subscribers. However, there's a simple alternative that provides the same functionality without the premium price tag.
Option 2: Building a Free Alternative with Airtable
You can create equivalent human-in-the-loop functionality using virtually any app that allows you to easily create, reference, and edit data.
For example, you can swap out the official “Human in the Loop” step with an Airtable step instead, and you can do it with Airtable's free tier.
Airtable is a no-code database app that works great as a companion to any automation provider. You can learn more and get started with our Airtable beginner’s guide.
Alternatively, you could use the Smartuite, Google Sheets, Notion, or any number of other apps.
The process will be very similar to the setup we covered for the Human in the Loop module, and will still involve two automations (or more, if you want to run automated actions for different decisions).
First automation: create and share a record in Airtable to review the AI output
1. Set up your trigger in Make
2. Send a request to an AI application
3. Create a new record in Airtable
After you’ve triggered your first automation and sent a request to the AI, you can store its response in an Airtable base.
You’ll want to create the database in Airtable before adding the module in Make. We’d recommend including the original request details, as well as a checkbox that you can use to indicate that the content is ready to be sent out.
Finally, you should add a new view to the table. Add a filter to the view, and set it to only include records that have been checked off. We’ll use this later to trigger the second automation.
Once your table is ready, just add a new Airtable module and select “Create a record” as the action. Then, map each piece of dynamic data to the appropriate Airtable field.
4. Send a notification (e.g., via Slack) with the Airtable record URL
Instead of sending a Human in the Loop URL, you can send out a link directly to the relevant Airtable record.
To construct this URL, just open up any record in your Airtable base (expand the details), and copy the full URL up until the string of characters that begins with “rec”. This string is the record’s unique ID, and we’ll replace it with the dynamically retrieved ID for each newly created record.
Paste this URL into your Slack message in Make (or whichever app you’re using to send the URL), and add the “ID” variable at the end.
Your first automation should now be all set. You can test it out to see a Slack message with a link to the latest record in Airtable.
Open it up to make sure everything looks correct, edit the AI prompt if you’d like, and check the box to get ready for the second automation.
Second automation: send the approved AI output to its destination
Your second automation will trigger whenever a new record enters the designated view for approved responses.
However, you don’t necessarily need to build this automation in Make. If you’re using Airtable, and want to send the AI’s output as an email or Slack message, we’d actually recommend building it there instead.
Automations in Airtable are free and don’t limit operations in the same way that Make does. If you have a scenario in Make checking for new records even just every fifteen minutes, you’d be using up 96 operations a day with the trigger alone. This would quickly use up your monthly allotment on lower tiers.
Airtable would avoid this issue, but does have limited integrations. If the app you want to send the AI output to isn’t supported in Airtable, you’ll have to use Make instead.
If you’re new to Airtable automation, you can learn how it works in the second part of our Airtable beginner’s guide.
The automation that finishes up this workflow will look like this:
1. Launch when a new record enters the specified view
2. Send the AI’s output wherever you want
In our Airtable example, we’ll send the AI output via Gmail.
With the completed workflow, an AI response is generated whenever a new row is added to our Google Sheet. The response is stored in Airtable, and sent to our team for review in Slack.
We can then edit and approve the response in Airtable, and automatically send it out via Gmail.
We’ve got quick and easy oversight in our AI automation, and we don’t need an enterprise plan in Make to do it.
Save money on AI automation with simple, DIY alternatives
Human-in-the-loop automation doesn't require enterprise software. By combining free tools like Airtable with your existing automation platform, you can create robust review processes that ensure quality while saving time. This approach gives you the best of both worlds: AI efficiency with human oversight.
If you’d like to learn more about automating your daily work with tools like Make and Airtable, be sure to explore the other posts on our blog or our YouTube channel. You can also follow XRay on Twitter, Facebook, or LinkedIn.
