I just don't buy this happening for software engineers. I feel AI will always need human supervision with a human guiding the AI who lacks the wider context that a human can gather. I have really enjoyed doing some coding recently with GPT on some front end stuff, which is where I suck (I am a backend / systems engineer). GPT taught me a lot, but it soon got itself in a lot of trouble and had I not intervened a bug ridden project would have been made. I realise the tech will improve, but I just can't see it faring well on its own without a human holding its hand and feeding it with the context that a machine cannot determine on its own. You can see this through where it is strong, writing unit tests - testing is a perfect fit, it can read the various types , functions and methods and determine what the outcome should be, and then write a test to check those outcomes are realised.
Rather than replacing software engineers, I see it being more of an enhancer.
Many engineers will be able to upgrade their productivity and find a lot of use from being able to bounce back and forth with GPT. This was my experience this week, I worked 'with the AI' and every time I got something working it felt like a shared high five moment.
Us being out of work? Nah, this sounds like the same sentiment before that calculators and then computers were going to make accounting a job of the past.
I think that this is a naive take. Current AI capabilities like ChatGPT can be compared to Pascal and C. We will get the Python and Java equivalent when we abstract things with hyperparameters and scale up the number of params at some point.
It's a really weird analogy because I think the only critical difference between C and Java is that the latter has garbage collection. What's the equivalence of garbage collection/memoery management in AI...?
which is why it's not a good analogy: you make a powerful tool like "programming language" easier to use, you get more people with the power to write programs. You make an unreliable and brittle tool like ChatGPT easier to use, you get more crap content on the Internet.
Not saying that there won't be good use-cases for Large Language Models, just saying that the analogy doesn't work: these things will be useful if used by competent people in very specific use-cases. The way ChatGPT is used now is just to generate a lot of hype and middlebrow content: with that in mind, I think its low barrier of entry is actually a negative (it causes things like this: https://www.npr.org/2023/02/24/1159286436/ai-chatbot-chatgpt...)
I started writing in Python this January, after spending years in Java, C, C++ ad PHP. (In academia, Pascal.)
I would say quite explicitly that Python is a lot better than C, at least as long as you value pure concentration on the business logic of whatever you are implementing. Its libraries are helpful, huge and well documented.
I am overjoyed by the fact that I don't have to a) either reinvent the wheel myself (e.g. writing a JSON parser from scratch) or b) rely on some OSS library with a bus factor of 1, whenever I just need to store or read data etc.
The Python honeymoon period lasts a long time. For me it lasted until the monolith I was working on had grown to ~500k lines of code. At that scale the smallest changes would cause (seemingly) unrelated breakage, despite many thousands of tests saying all was good. Python’s “anything is anything” philosophy made me miss the predictability of C and Java. Granted, that was before type annotations had taken off so maybe things are better nowadays.
In that case, blessed is me, because my Python projects are small-ish mutually independent modules ("probes") that only communicate using a specific JSON-based protocol over HTTPS.
I should elaborate that when I write "better than nothing", I'm someone with a background in strongly-typed programming languages (OCaml, Haskell, Rust) and even more strongly-typed programming languages (Idris, Coq, Twelf).
It's unlikely that Python type annotations will ever be able to achieve something comparable to what many of these languages had during the 90s. They're useful, but if you need strong static safety guarantees, Python is the wrong tool for the task.
If AI makes engineers more productive, then it follows that companies will hire fewer engineers for the same work. So AI is making some engineers obsolete.
On the flipside, it might create more demand for some engineers (or skills in general) as the bottlenecks move "upward" on the skills ladder (if we define that ladder as "how hard is it for AI").
E.g., faster and cheaper compute created more demand for software developers as that demand was no longer capped by the compute bottlenecks. Similarly, faster and cheaper "basic" software development might create more demand for software architects, and so on.
> If AI makes engineers more productive, then it follows that companies will hire fewer engineers for the same work.
History says otherwise. When I started (late 90s I guess) we looked at starting a company. It was going to require things like big iron, rented DC space, teams of people to manage, for what was something that is now pretty simple. Fast forward and the same idea could be done (and has been done) by a couple people using a cheap VPS.
Every productivity enhancer that has happened over my career has not led to fewer jobs, but more. The ability for humanity to consume any excess resource is unmatched. I'm sure there will be some bumps along the way, but any company who thinks they can stay static and replace everyone with AI will get beat out by the companies using AI as a tool to continue to innovate.
Field is evolving for a long time, with more automation/frameworks which make programmers more productive. So far it increased workload and opportunities.
The real issue will happen if suddenly humans will be completely removed from the loop and AI will scale independently.
If AI makes engineers more productive, then it follows that the application of engineering become economically viable on whole new sets of problems. So AI is making engineers even more sought after.
That said, I'm so unimpressed with AI at the moment that I start thinking it's just a bubble, but even if it isen't, it's not going to lessen the need for engieers in the least.
> then it follows that the application of engineering become economically viable on whole new sets of problems.
Agree.
The demand curve for highly skilled SWEs is incredibly robust but has been supply limited. Insofar as AI effectively upskills workers who were just below the demand cliff, this is a boon to workers.
The big shift I see is towards Quality. Implementation has gotten cheaper relative to problem specification and verification.
I don't see why these are mutually exclusive. The point is that those new "engineers" will not necessarily be the same people. Those "good" engineers will enjoy improved productivity and will be more sought after, while those "not so good" engineers will be replaced by AI. As always, any innovation just seems to increase inequality.
On the contrary, AI will _also_ create jobs for people who as of now would be unqualified for engineering.. People who have enough of the right type of thinking, but can't learn to code well enough to be productive, they can now be AI babysitters and integrators.
chatGPT has proven a productivity booster for me. I had to port some typescript backend code to google go and with chatGPT it felt like I had access to an experienced google go developer to ask questions to.
For juniors it is a massive booster as they have someone available all the time allowing them to explore further before asking for help.
If the whizz-kid and his fancy AI make you get wrong results that may bankrupt you or make you end up in jail, well you may have been better of with the abacus. But as always the temptation if too high; for some hope is indeed a strategy
Your belief isn't required. After decades of paying a kings ransom to twitchy weirdos who sandbag deadlines, blow through project budgets like a freight train through tissue paper, and who's final product has a nasty tendency to cause as many problems as it solves, business managers are thoroughly incentivised to explore alternatives.
This take is very yikes. Your assumtion that all coders are twitchy weirdos reflects back on your own distorted assesment of a whole industry probably based on few personal anecdotes. But if you hope AI will make things smoother for you you’re in for a rude awakening
Idgaf if you think it's yikes, it's based on 25 years of working in and near the industry as a developer and listening to whats said when folks think there are only suits in the room. If you think D and C level managers/execs love their software guys you are deeply delusional.
Rather than replacing software engineers, I see it being more of an enhancer.
Many engineers will be able to upgrade their productivity and find a lot of use from being able to bounce back and forth with GPT. This was my experience this week, I worked 'with the AI' and every time I got something working it felt like a shared high five moment.
Us being out of work? Nah, this sounds like the same sentiment before that calculators and then computers were going to make accounting a job of the past.