How will AI change how Salesforce is delivered?

In the mid-1990s, I read a book by Ed Yourdon with a quote that made a point that stuck with me (even if the exact quote did not). The quote was something to the effect that “technology changes more quickly than we expect, and people change less quickly than we expect.” In the latter part of the 1990s, Ed Yourdon went on to write a series of books about Y2K. In the late 90s, a friend of mine, Ed Meagher, who hosted a radio show called “Y2K Today” about, not surprisingly, Y2K, was talking to Ed Yourdon and mentioned this quote. Ed Yourdon said he’d been quoting someone else and shared the person’s name.

Unfortunately, in the intervening two+ decades, I have forgotten the exact quote and the original source, so I can’t appropriately attribute the quote.

The gaps in my memory notwithstanding, I still think this is a great quote: “technology changes more quickly than we expect, and people change less quickly than we expect.”

A common theme in economics and business is the idea that when the cost of a resource becomes very low, there will be changes in the way the resource is used as well as changes in the businesses in the delivery ecosystems that include the resource. The more abruptly the resource cost drops, the more rapidly the affected businesses need to react. The businesses that react correctly and quickly are the businesses that will experience game-changing success.

Combining the above with what has happened with AI in 2022, we have:

1. A technology, AI, that has hit an inflection point in the capabilities it makes available to the general public,
2. A resource, intelligence, that has just had its cost drop by several orders of magnitude,
3. Business ecosystems that leverage this resource need to react to the cost drop, and
4. As the person Ed Yourdon quoted in his book noted, “people change less quickly than we expect.”

The AI-driven change coming to Salesforce, Customer Relationship Management, and all facets of business activities that use intelligence suggests we’re headed into a period of both chaos and of opportunity.

Staying focused on Salesforce for the moment, I first deployed Salesforce in 2002. Back in the day, to edit a page layout required writing down what needed to be changed and then clicking into the admin pages and navigating to where the page layout could be edited. I used to say that the most impactful change in Salesforce for development was when Salesforce added the ability to edit the page layout right from the page to be edited. With what I have seen AI do across 2022, I suppose I need to stop talking about editing page layouts and start talking about the impact of AI on Salesforce development.

AI for Salesforce development – configuration, metadata, Apex, HTML, CSS, JSON, and more – is nothing short of amazing.

To write a Call Out to send Account data to other enterprise systems when an Account record is updated is typically ten (10) hours of work. Once the requirements are defined, using Scrum the process is as follows:

1. Write a user story,
2. Groom the user story,
3. Get the product owner to sign off on the user story meeting the Definition of Ready,
4. Have the Solution Architect or Lead Developer add design notes,
5. Assign the user story to a sprint,
6. During sprint planning, assign the user story to a developer and answer any questions they have,
7. Have the developer do the development,
8. Have the developer unit test what they developed,
9. Have what the developer developed be peer-reviewed, and
10. Send the work to the QA team.

With AI, the same process is less than one (1) hour of work. With the requirements in hand, the process is as follows:

1. Write an AI prompt for the JSON,
2. Write an AI prompt for the Apex Call Out using the JSON, and note – the JSON doesn’t have to be cut-and-pasted into the prompt for the Call Out; the prompt for the Call Out just needs to include “using the JSON you just created,”
3. Write an AI prompt for a test class for the Call Out,
4. Write an AI prompt for a comment header for the Call Out,
5. Paste what the AI has generated into a sandbox,
6. Unit test and “peer review” what the AI-generated, and
7. Send the work to the QA team

Please note – I didn’t say “the process will be” but rather “the process is.” Tools like OpenAI’s ChatGPT do, today, steps 1 through 4 in the list immediately above.

All of the above means that the move to AI for Salesforce delivery is not just about continuing to use the same processes we use today but will also require taking a step back and looking at the delivery methodologies we use. With a little bit of automation to combine steps 1 through 5 into a single prompt and an AI to do the unit and QA testing (which isn’t available at the moment, but I expect someone to release before the end of 2023), tranches of Salesforce development become so inexpensive that today’s delivery methodology overhead becomes price prohibitive.

We need to think about how Salesforce will be delivered in a world where in a discovery session a requirement for sending updated Salesforce account data to the ERP system is surfaced and while the business analyst continues to elicit requirements from the stakeholders during the meeting, the Salesforce and ERP architects sitting in the back of the room use AI to build the production-ready integration. Well, perhaps the “Systems Integration Testing (SIT) ready” integration, anyway.

Will we find a way to use existing methodologies – waterfall, agile, Scrum, Scaled Agile, Team Topologies, etc.? Or will a significant evolution of how we do Salesforce delivery be needed?

If you have thoughts on this, I’d love to hear them!