Last month’s release of beta AI features into Agilebase prompted the most great feedback we’ve had in a while. People were excited by the possibilities:

• “Fascinating to read about this”
• “It’s fab to see AI being introduced to AgileBase”
• “I gave it a test run yesterday when I created a new table and from one sentence it was able to connect my new table to an existing one in the correct way and make some very good suggestions in fields to use.”
• “Looking forward to what more AI can bring!”

A lot of ideas were sparked, too:

• “Just thinking it would be very handy to have maybe a direct built in assistant when submitting calculations, so something that could highlight when you go wrong and submit an incorrect effort and suggest you a correct version and guide where you went wrong. I think this would be very very useful for new and existing users.”
• “It would be really good to have this as an option when creating calculated fields – I’ve actually already been using chatGPT 3.5 to help me with some more complex calculated fields, since my knowledge of SQL is very basic”
• “After it creates the table, asks if you’re happy with the result, if not, allows you to input your request again and re-does everything again (including disconnecting any connections it made)”
• “Users having a quicker means to have custom ‘user unique’ views would be amazing. Would save me a lot of time!”

and one important concern:

• “We’re actually now starting to look at AI use internally and one of the major concerns is that giving AI access to sensitive information could accidentally leak this into external servers and be open to unauthorised access.”

We’re pleased to report that many of the above ideas have now been implemented! However, to address the data privacy question first, the third party AI service that we currently use is OpenAI’s ChatGPT. They have a set of policies here: https://openai.com/policies

We have now added links to these policies from our documentation, e.g. https://docs.agilebase.co.uk/docs/artificial-intelligence/

We strongly suggest that people read the data privacy policy. For UK/EU residents, make sure to click on that version.

At the time of writing, OpenAI’s policy states that “We do not use […] API data, inputs, and outputs for training our models”.

Note this is in contrast to when you use ChatGPT manually, through the browser, when they can retain and use the text you type. 

For clarity Agilebase uses the API. However, as with any third party service, policies can change. If your organisation believes this might cause a commercial risk, please consider the data sent, as outlined in the documentation above.

We have added an entry into our ideas and issues log to look at ways to inform users of policies and potentially control access to features at the organisational level. If you’d like to be part of this conversation please do get in touch.

What’s new?

Now, onto the new features following user feedback:

1. When the AI suggests fields to add to a table, they are now not added immediately, but presented as a list, from which you can choose fields to accept or reject. If you ‘reject all’, you can also change the prompt that you gave and get it to try again, i.e. work iteratively to arrive at the best solution.
2. The AI can now create calculations which will be a game changer for people who don’t know SQL. The goal of Agilebase No Code is to empower non-technical people so this fits right in. Please note that the AI might not always get them right! It’s not perfect, however it should be a good learning tool. 
3. If you enter your own calculation and it has a mistake, the AI will suggest possible ways in which you could correct it

Here’s an example from one early test our CEO did – to calculate the financial period which a particular date falls into:

What’s coming?

As well as continuing work on making development of applications easier, we’re also planning on making the power of AI available for you to use on your own data as well. For example, you could ask it to look through all your active sales enquiries and suggest next steps for each prospect.

This feature is likely to have a cost associated. We will try to make this as simple and affordable as possible. We use the most up to date (and higher cost) AI model, as our testing has found that it produces significantly better results.

I’m sure you have ideas for how this could be used. Please contact us and let us know what they are, so we can ensure that the way we design our features matches your needs.