Poll: Is coding still a viable profession?

[willendure]

This is something I've been wrangling with as my son is looking at applying next year for either a Computer Science and/or Maths degree.

It's a very mixed picture for recent Computer Science and Computer Engineering grads though.

A recent graduate labour market survey in the US found of 74 degree subjects, Computer Science had the 7th highest unemployment rate, Computer Engineering 3rd highest and Information Systems Management 9th highest. So 3 IT related degrees in the worst 10 for unemployment.

The flip side is those that did manage to find a job where more likely to find a degree level one. All three subjects had much lower than average 'underemployment' rates. Median salaries for early career grads (under 27YO) were still very good) but I suspect are flatlining or declining now.

This is consistent with what I've been hearing anecdotally from UK IT recruiters.

In short, it's probably still a good degree but make sure you go to a decent uni and get some related internship or other work placement experience.

There is something about that on the H1B Visas thread. In the linked article: https://www.theregister.com/2025/07/...b_job_lottery/

"In 2023, American colleges graduated 134,153 citizens or green card holders with bachelor's or master's degrees in computer science," Lynn wrote.

"That same year, our federal government handed out work permits to at least 110,098 foreign workers in computer occupations through just three major guest worker programs. That's equal to 82 percent of our graduating class who are guaranteed jobs even before any Americans walk across the stage for their diploma."

It seems like multiple factors at play - cheaper labour undercutting home grown, a glut of graduates since the profession was well paid and a lot of people wanted to get in, post covid lay-offs in tech, end of cheap money... The dynamics are hard to interpret as to whether this is a cyclic thing or more permanent shift.

Your son could consider also Engineering or Physics or AI? My boy will be making deicisons around higher education in the next year or so with similar choices to make.

[edison]

I will repeat my default advice here - Degree Apprenticeship first (and the Civil service do a lot of very good ones) and University second.

Sounds great in theory and I tend to agree but there simply aren't enough degree apprenticeships to go round and the best are extremely competitive.

A lot of the quality of teaching leaves something to be desired and often with low ranked universities and colleges. The best bit about most apprenticeships is learning on the job (as by definition, it should be.)

I'm not sure about degree level apprenticeships but across all apprenticeships, the dropout rate is something like 35-40%.

[NotAllThere]

... if you give it a tight spec...

Any program is a specification, just very precise. Since you need to give an AI a tight spec, effectively the spec becomes the programming language. People who are crap at coding won't be able specify well enough. So the idea that execs will simply tell an AI what they want and it'll go away and code it is just fantasy. At least for the next ten years.

[WTFH]

Any program is a specification, just very precise. Since you need to give an AI a tight spec, effectively the spec becomes the programming language. People who are crap at coding won't be able specify well enough. So the idea that execs will simply tell an AI what they want and it'll go away and code it is just fantasy. At least for the next ten years.

The idea that execs/management could come up with a precise spec is never going to happen. That would require intelligent management.

It reminds me of a time, must be about 20 years ago, where I was contracting in to a company who outsourced all their report development to an offshore company. One report that was required in the UK needed to output in GBP based on a figure that was calculated in EUR. In my spec I had mocked up the output, and shown the fields etc. During one of the calls, I said to the developer "take the value in Euros and convert it to pounds". That was precisely what they did. They converted the value to pounds. Not pounds and pence, just rounded it to the nearest £1. It didn't matter what the spec said, or what the mock-up showed, when I said "pounds" and meant currency, they heard "pounds and nothing after the decimal". They lacked the ability to apply logic to what was said, or to consider all the other related information.
But it was my fault for not making it clear.

[jamesbrown]

But it was my fault for not making it clear.

It could've been worse, they might have entered the weight in lbs of one euro coins.

[willendure]

Any good developer knows that when you are dealing with currencies you must use a Decimal implementation, and give consideration to the number of decimal places needed and how rounding is handled.

[willendure]

Any program is a specification, just very precise. Since you need to give an AI a tight spec, effectively the spec becomes the programming language. People who are crap at coding won't be able specify well enough. So the idea that execs will simply tell an AI what they want and it'll go away and code it is just fantasy. At least for the next ten years.

Funny. Current PM has failed to organize the business requirements and high level design for our next piece of work, which means that the schedule to complete it is now very tight. He threatened the dev team with having to take the work to an external supplier in order to get it done in time. Obviously the first thing the external supplier is going to ask for is the f'ing spec, which is the whole reason for the delay. Jesus wept.

[willendure]

I would say though, that AI can be very useful in writing requirements docs and technical specs. I have one GPT that I found called "Cloud Architect Pro" which has been absolultely spot-on for generating accurate design recommendations for AWS. It has output examples for me in CDK or Terraform, or written them up in English for inclusion in technical design docs. Huge time saver.

The way I see gen AI is its like an extension of your left brain. The left hemisphere is more concerned with tool oriented thinking, or fulfilling tasks, language and use of hands and its reward mechanism is dopamine driven. The right side is more concerned with awareness and pattern matching and can deal with multiple conflicting ideas at once and its reward mechanism is more adrenachrome or some kind of adrenaline derivative. Left side gets dopamine fatigue and right side does not.

AI is not conscious or aware, but it is getting good at the language/tool-oriented stuff.

[hobnob]

Any program is a specification, just very precise. Since you need to give an AI a tight spec, effectively the spec becomes the programming language. People who are crap at coding won't be able specify well enough. So the idea that execs will simply tell an AI what they want and it'll go away and code it is just fantasy. At least for the next ten years.

In my CompSci undergrad degree, I did a module about formal specification in Z. The theory was that you could define what the program was supposed to do, rather than how it was supposed to do it.

E.g. the Z spec for a square root function would be something like this:

sqrt(x) = y, where y * y = x

(Not actual syntax.)

The implementation would be a couple of pages long, using something like the Newton-Raphson method.

However, I've never come across anyone using that in business. I heard about one (possibly apocryphal) case where the military decided to use a Z spec for their missile guidance system. The result was a shelf full of bound volumes, and the general basically said "How do you expect me to approve that?"

I agree with NAT that the vibe coders are unlikely to give a rigorous spec to the AI.

[willendure]

These specification languages are not mainstream but I have heard stories of people using them with gen AI - describing scenarios in words and having AI turn that into formal specifications. Alloy executable specifications might be a good one for testing generated code against.