Friday 17 March 2023

Einstein GPT - Rise of the Machines?

The very first computer magazine that I bought, way back in February 1981, was an issue of Personal Computer World with a story entitled "THE LAST ONE". This was a revolutionary system where you described what you wanted the program to do in plain English and it generated the code for you. This was being sensationalised as the end of programmers as we know them, yet I've managed a fruitful career in the 4 decades since.

Plain English was doing a lot of heavy lifting here, and the reality was that you chose options from menus to build a flowchart that eventually generated BASIC code, but it was pretty cool for it's time. Fast forward 42 years and with Salesforce announcing Einstein GPT at TrailblazerDX 23, it appears we'll be able to describe our requirements in a plain text comment and the class will be generated for us. So does this finally mean the end of programmers as we know them? Once again I think that would be a premature conclusion to jump to. 

If you'll permit me another stroll down memory lane, I'll use a documentary on factory automation to explain why. This was a ceramics factory that produced, among other things, mugs. It was the job of the final worker on the assembly line to take the mug and place it in a cardboard box for shipping. The company had invested in automation and he'd been put on notice that he was to be made redundant as part of this push to automate all the things. The slight downside was that the robot that picked the mug off the assembly line and put it in the box wasn't initially very accurate. it was knocking mugs off the conveyor when it tried to pick them up, and often dropped them when swinging round to the cardboard box, resulting in huge breakage numbers. This worker was then retained to stand near the robot and catch the items that it knocked over or dropped! The robot was continually being adjusted and adapted to do a better job, so he was essentially acting as it's minder until it got smart enough not to need him, at which point he'd be let go.

I think that we are in the same situation with Einstein GPT (and all the other GPTs) - we can ask them to create something, they will have a go, but you still need a human there to check that they got it right and fix what they got wrong. For simple classes with a method or two, it will make a good fist of it, and the same for the test classes, but as the complexity rises it's likely to need more human intervention. Human guardrails if you will, ensuring things stay heading in the right direction.

What Einstein GPT will give us is very useful starting points, rather like templates. Just like templates of old this will remove the need for us to continually rewrite the same few lines of boilerplate code to declare a class, except that these templates will be ever evolving based on what is already present in our codebase, and generated on demand based on what we can articulate (for regular classes) or what a class currently looks like (for test classes). For simple requirements they may be a very good first cut that just needs a bit of polishing.

A real time and money saver if most of the code written is relatively straightforward, and a potentially huge productivity boost for junior developers, as long as it gets it right most of the time. Experienced developers should also get a productivity improvement - they'll focus more of their time on the hard stuff rather than setting up the basics.  

So will Einstein GPT take the jobs of Salesforce developers? I think not. 

Salesforce Developers that use Einstein GPT may take the jobs of Salesforce Developers who don't, especially if the majority of the development work is low complexity.




No comments:

Post a Comment