The world is abuzz with the mind-blowing capabilities of the recently released public beta of ChatGPT-3, a prototype artificial intelligence chatbot developed by OpenAI that focuses on usability and dialogue. Countless examples of its extraordinary capabilities have been shared on social media showing ChatGPT developing movie plots, writing songs, and creating recipes based on user input.
At MXA Solutions, we have been both surprised and impressed with ChatGPT’s knowledge of Salesforce which raises questions about the future of Salesforce consulting. For example:
• What will Salesforce customers expect from consultants with tools that make customizations seemingly easier?
• What will Salesforce do with AI for customers who require customizations?
• When development is commoditized, what is the role of the consultant?
• How will Salesforce itself change?
These are existential questions for not only the Salesforce consulting world, but for everyone in the Salesforce ecosystem. In other industries, such as language translation, AI became a cheap alternative to experienced translators. The quality of translation versus the cost of translation became a trade-off businesses had to make. For translators, working with AI translations may have meant spending more time editing and sometimes reworking from the original source.
A similar future may await the Salesforce consulting world. How far away is that future? Determining the exact timeline is difficult, but the pace of AI interest and financial investments would suggest it may be sooner than we all think. In the near term, when using artificial intelligence tools like ChatGPT, we will need to apply common sense and take a measured approach. At MXA Solutions, we are optimistic about how AI can be a helpful tool in how we deliver value to our clients and have been putting ChatGPT to the test.
For example, we recently had a requirement to develop a validation that could be completed using a trigger:
In 5 seconds, ChatGPT scaffolded a basic trigger that could be copy and pasted and even added comments. However, did you spot the glaring issue with the trigger? We asked ChatGPT about the issue, but purposely left our question vague:
And asked again, this time more explicitly about the glaring issue:
We asked again, but challenged some assumptions about how this should be implemented:
Woah! A SOQL statement inside of a validation rule? We pressed the ChatGPT about this and it insisted this was possible, but we have yet to confirm that this works.
Given the experience we highlighted above, our takeaways are as follows:
• In its nascence, ChatGPT is already an incredibly useful tool to help draft ideas into actionable steps. The answers it provides could be 80% of what you need, but it is not a finished product.
• The more targeted your inputs, the more precise the outputs. Much like conducting a Google search, you need to know how to ask the right question. In our example, we were specific about a trigger. However, what if the problem only needed a validation rule? Asking the right questions is important.
• In many ways, ChatGPT is like a contractor who will just take your order and do exactly what you tell it to do, without applying any thought or asking any questions. You still have to do some thinking about business and systems context to ask the right question for the AI to deliver a meaningful answer. Otherwise, it will do what you literally ask it for.
• The future of declarative tools will be interesting. Will Salesforce make tools available to customers and developers to accelerate development and configuration?
Experience and familiarity with Salesforce are still irreplaceable. Even with the billions of bytes of data that trained ChatGPT, it made one of the most basic coding errors every Salesforce developer should know about. AI will get smarter, but buyer beware for every answer you receive.
Subscribe or follow us to receive a notification when MXA Solutions publishes more great Salesforce content.