ChatGPT Isn’t Going to End Your Software Dev Job
ChatGPT and CoPilot might reduce the number of developers needed at some point, but larger forces are at play
I spent (wasted) too much time in my thirties wondering if I was going to get aged out of software development before THEN wondering if I was going to even make it that far! *
Since then I’ve gotten a chance to play with ChatGPT, CoPilot (in the form of Visual Studio Intellicode) and read an excellent book on the topic, PAIR programming -which isn’t about what the title suggests).
I have seen shocking predictions about how many jobs will evaporate during the AI revolution. Silicon Valley and Venture Capitalists are not holding back from exploring the new, generative frontiers, whatever distruptions lie in their wake.
ChatGPT itself claims 5 million people could lose their jobs. [1]
If you’re wondering whether everyone is going to abandon the software development market (giving you the chance to sail with the remaining software work), think again. College CS major enrollment is at a high and growing. [2]
I checked out ChatGPT and CoPilot for the first time last year, and got mixed results.
On the one hand it seemed like half the time it knew exactly what I was going to type for a given line. The other half of the time it seemed to give me nothing, or it was way off.
I’ve since learned after reading Pair Programming that you can provoke AI generated content by putting in comments and that it’s normal to delete and regenerate content.
There were a couple times where I would just name a C# method, and IntelliCode (i.e. CoPilot) would auto generate the entire method for me! I was blown away.
But could that kind of capability make it hard to justify paying a software developer enough money to support a house and a family ? My doomsday scenario is getting my first born enrolled in an expensive, famous college and then becoming unable to have any kind of salary.
A friend recently told me her husband generated a shopping cart using ChatGPT. A process that used to take him a week (I’ve done it before and it took me a week), could be done in seconds with ChatGPT.
It’s conceivably possible that a company might choose to simply forgo hiring developers and just fulfill their software needs by using ChatGPT entirely.
And to that end, I’ll note that Microsoft recently decided to not raise the salaries of their software engineers. [3] This seemed conspicuous to me because it targeted software engineers and didn’t seem to be just dialing staff in general back to pre-COVID levels.
Here are some reasons why I don’t think generative AI is going to significantly reduce the demand for software developers:
Too Much Noise
In my own experience and what I’ve read from others, a lot of the AI generated content is off by a little or sometimes completely hallucinated [4]. While this hit-or-miss approach might work in the hands of a developer, managers are not going to want half of the submissions to be as faulty as they currently are. And the reductions in Quality Assurance staff are only going to lead to more users discovering the worst atrocities.
Outside of just coding, ChatGPT has become somewhat infamous for hallucinating court cases and other details that simply never happened. In the code it generates, it makes similar references to library methods that simply don’t exist.
This might get somewhat better over time, so stay tuned.
Cost
Training an AI and responding to all the potential queries from all developers asking questions is computationally intensive.
As in, a lot more intensive than a regular search engine search.
Security
A lot of companies are very anxious about their software leaving their control (many of them are corps or government off the shelf that are not wanted at all).
While boards and CIOs have swallowed hard and adopted cloud computing, many are still housing their data and computing on site.
The same trust concerns exist for ChatGPT and CoPilot which airs at least some information to the web.
Creativity
Michael Callaghan lists creativity as a liability for generative AI. Some software tasks require creativity (think game design or streaming content, where learning from and parity from other shows / games is just going to end up in mediocrity) and others not so much (tax prep, form entry, etc).
Growing Demand
Demand for apps is growing and not expected to plateau any time soon.
A lot of this stems from increased regulation and the general immaturity of the industry.
Deprecated Information
AI programming tools are generally not aware of any software updates within the last two years (Callaghan). The same dated information applies to ChatGPT as well.
In fields such as behavioral psychology or masonry, this might be an acceptable delay, but given the continuous learning required to write software, this is a major setback.
Imagine hiring a software developer who has not learned a single thing in the last two years. This might change at some point, but at the moment, AI is going to be well behind developers who are staying current.
Lack of Personal Insight
Generative AI doesn’t see the human cost or human value behind design choices or implementations.
While an AI trained on existing interfaces may understand that alerting users to a widespread emergency crisis should not be a default action, new functionality outside of exist
Declining Demand ?
Interest in ChatGPT and CoPilot may already be on the decline.
When Microsoft announced generative AI access on Bing, their traffic started to rocket up (as Alphabet shares began to decline).
But now traffic to Bing is on the decline [4], which may indicate the near term disruption from AI produced content may have already peaked.
Summary
Software developers may continue to feel the changes resulting from the advent of generative AI.
As others have noted, the need for understanding the underlying business and high level design of software is not likely to be replaced by ChatGPT or CoPilot any time soon where as those who see themselves as “coders” might have a bigger time finding work.
While AI is expected to grow in capability and usage over time, it isn’t looking like it will be a widespread substitution for human software developers any time soon.
*Age discrimination is certainly nothing to sneer at. It’s the single, widest form of job discrimination both inside and outside of software development. Of course corporate training videos never address the true victims of discrimination (Cubans, business owners, and, here, the elderly).