Am I still relevant as a programmer now that there's AI?
andrewhuth.substack.com
andrewhuth.substack.com
What kind of stuff do you work on?
It has become much easier to do stuff like that even if you suck at photo editing.
When you pay a person, you're paying for more than the output. Maybe they'll use a GPT/LLM model behind the scenes, but I also use bucket-loads of coffee behind the scenes. That doesn't mean a bag of Stumptown can build your analytics architecture.
As an example of this I generated an image of a dog with an AI and then I put it in the vectormagic.com's tool. The result is very good since the image is not a logo style but an advanced image with a lot of detail.
https://vectormagic.com/images/kvzrgcxyjp45m/edit/p24qfcqegj...
So yeah, still stands. If you can stomach messy, unpredictable outputs, use an automated solution. If you want something lean and usable, use a human-being.
In my career I’ve heard about one thing after another that was going to make programmers obsolete. It may happen but I doubt it, because coding isn’t the hard part of software development.
Perhaps people who nowadays make a living on making small websites for small mom- and pop-shops will replace themselves with an AI and they will mostly care about design instead. They will have more time on other stuff, increasing the quality all over.
That said, it's hard to tell either way. The future is hard to predict, I wouldn't have guessed I could generate the amount we now can just one year ago. It's hard to imagine many years of progress and what that could potentially lead to.
Why does anyone doing that kind of work still get jobs? Ignorance plays some part, of course. In my experience a lot of business owners believe they are unique and special and need something “custom.” It’s a status thing as best I can tell.
Back in the ‘90s I started refusing to write custom accounting systems for customers because QuickBooks could do everything they needed until they graduated to the big leagues, where enterprise software solutions dominate. They never listened, always sure they were too different to use something as pedestrian as accounting software in a box.
Programmers aren’t immune. Look at all of the functionally equivalent languages and tools we argue about. Notice programmers writing that they can’t use an IDE or web site that doesn’t have dark mode. How many more “to do list” apps will we endure? Freud called it “the narcissism of small differences.”
The same human tendency explains why we have many more cars available than Toyota Camrys and Ford F-150s.
Writing the actual code once it gets to it is usually not the hard part, unless you work on particularly difficult problems. LLMs help but don't replace us just yet.
Programmers who focus all their efforts on skills that can get automated will eventually get automated away. When a chatbot can replace me in terms of understanding customers and business domain expertise, and translate requirements into code, I’ll retire.
With the correct plugin, yes.
I see this being at least a few years out. Performance is stil scaling up and cost is still scaling down. It’s going to require collecting a lot more specialized training data to produce fine tuned model. Business will need to adapt and people will need to learn how to accelerate themselves. But eventually, I see this going the way of the introduction of the computer to the office. No productive person will work without it.
As a thought exercise, I’m going to assume you’ve used same GPT model to produce some code before and been impressed in a narrow kind of way. Then project that forward a few years, through GPT-4 and GPT-5. Add some better training data and a better understanding of how to fit the LLM output into day-to-day work. Maybe add the ability to fine-tune on the existing code base and design documents. The trend-line is at least going to hit junior dev level, so the job will probably turn into a senior dev supervising a “team” of LLMs.
I think that same accessibility is going to be a major driver for these natural language models. I’ve seen great examples where people can basically “program “ in English by having a conversation back and forth with the LLM and refining its output of code.
Of course, there will be companies who'll get the one bright-eyed young person whom they can exploit the most, and get them to apply ChatGPT to build all of their systems, cheerfully reimplementing all the most popular architecture flaws and (e.g.) security issues that are seen in the many repositories of old code used in training the model.
Programming is a first order thinking, a way to deconstruct and construct things, sometimes you have to use it on a computer, sometimes you have to use it while baking bread, or while woodworking.
AI will help me so much, I am super excited, and now I use it every day, I plugged in whisper+chatgpt[1] to my emacs and now it is more and more like a partner, it is of course still stupid, but it will get better, and soon (with the recent llama work people are doing) I hope we will be able to finetune it.
We are doing lists and tables and data transformations and input validation for 60 years now, we built gazillion specialised spreadsheet clones.. I think now we can start building some new things, some really "soft" software, where the users can program it with language.
This is the first time I feel the computer is helping me do something, and its not me "fighting" with it.
but i think we are talking about the same thing :) programming is creative, and for me it is art, i love it.
https://github.com/facebook/react-native/issues?q=is%3Aissue...
People miss understand (and most business failure to realize the potential) of IT in general which is not typing functions in the latest programming lang...
It is problem solving. The number if times I have sat down with a business user to go over a problem they are having only to have to unpack the Rube Goldberg machine of illogical processes, datasets, and dependencies they have created to arrive at their "solution" they are now asking for my assistance with.....
I dont look for AI, at least not ChatGPT AI to solve that. Most IT problems are Solution XY issues. ChatGPT can solve to Y, but understanding X is the real trick
I was working with really, really smart PHDs and was doubting my hiring to the team. I felt like a lot of these guys had forgotten more than I'd ever known.
During my time in that job, I realized that my experience in navigating the complex databases from working with the business and operational teams gave me an advantage. I took my domain knowledge for granted and was able to work on the problems at hand much faster than the highly intelligent colleagues who relied on IT/DBAs to write their queries. They often had to go back and forth for a couple of days to clarify their requests, leading to misunderstandings and delays.
In the end, I fit in just fine, held my own, and was aware of my individual talents. I enjoyed learning from the PhDs, and they were happy to teach from their backgrounds.
A lot of the experience from the big bank has 0 transferability to any other company. What made me stand out was I knew which tables of the gigantic data environment were the best to use, I had hundreds of already built queries for many different problems and I was dependable. It got me really far before I went off on my own.
That's one of the beauties of IT work - for almost all of us, unless you shoot your own foot continuously, maybe 95% of our skillsets are easily transferrable not only to another employer, but also to another domain.
Of course if you willingly position yourself to some tiny niche that makes you unfireable or just follow your passions unchecked, you may end up in similar situation... an advice, for an easy life don't do that, rather accept potentially slightly lower pay and enjoy life more, its anyway damn too short
In case it's not obvious, this is sarcasm. It doesn't work like that, nor is it likely to anytime soon. And even if that day comes in our lifetime (which I doubt), your boss or PM still couldn't just delegate tickets to the AI. The AI would need people with advanced knowledge of both the AI and the domain who understand how to supervise it, and that would most likely be you and me.
The tools we use today are incredibly powerful and time-saving without AI. I can whip up a reasonably interesting mobile app with a data backend I could scale to millions of users in less than a week and basically for free. And yet there are more programmers than ever. Which indicates that maybe these tools empower human engineers instead of cutting them out of the loop.
Although just as software engineering is very different today than it was 20 or 30 years ago, in 20 or 30 years it will be very different…
It's not actually very different than 20 or 30 years ago. The platforms are rather different, the end product is different too in some cases, but the development work itself is surprisingly similar. That doesn't mean there cannot ever be a real leap forward of course.
Then other days, I use chatgpt or bing and I see some silly response, or I remember my control theory background and I feel at ease again.
It’s a really horrible rollercoaster ride mentally.
Turns out he was kind of right and kind of wrong. On the one hand most programmers today know virtually nothing about computers and programming in the sense that a programmer in early 60s knew computers and programming. Yet programmers today are more prolific and in demand than ever before. I suspect we'll see something similar play out.
- lot of dev work is just plumbing, config files, packages, APIs and tedious non challenging stuff
- demanding roles have high barrier: leetcode, certifications, open source, github, networking and so on - it will get more imo
- coding is treated as a basic skill taught in schools, like english or maths - not itself a differentiator
AI still does not have access to most businesses internal context, terminologies and so on - that is where we still have leverage.
For a long time, the answer was, "No jobs are at risk, AI can't compete in any scenarios. At best, it's a tool."
Now the answer is, "Only a few jobs are at risk, AI can only compete in a small range of tasks."
It's possible we're at the beginning of a hockey stick graph.
So what would it look like for AI to make the leap to mid level developer? It would have to understand:
1.) The codebase
2.) The technical requirements (amount of traffic served, latency target)
3.) The parameters (must have code coverage, this team doesn't integration test, must provide a QA plan, all new infrastructure must be in Terraform)
4.) The end goal of some task (e.g. integrate with snail mail provider to send a customer snail mail on checkout attempt if it was denied for credit reasons)
It would then have to make a design based as much as possible on the existing code style and library choices and follow it.
This is all probably possible now, although perhaps not for a general AI or LLM. But someone could build a program leveraging an LLM to provide a decent stab at this for a given language ecosystem.
The hard parts:
Point 2 requires an understanding of performance which is a quantifiable thing, and LLMs up until now have been bad at making math-based inferences.
Point 3 requires the bot to either provide opinions for you (inflexible) or to be very configurable for your team's needs (takes longer to develop).
Point 4 requires a _current_ understanding of libraries, or the ability to search for them and make decisions as to the best ones for the job.
-----
What about extending the above for a senior role? Now the bot has to understand business context, technical debt (does technical debt even exist in a world where bots are doing the programming?), and other "situational factors" and synthesize them into a plan of action, then kick off as many "mid level bot" processes as necessary to execute the plan of action.
The hard parts:
Current LLMs are pretty uninspired when suggesting ideas.
Business context + feature decisions often involve math, which again LLMs aren't great at.
But then we'd have to coerce the bot into generating structured responses that can act as next steps.
What exceptions should I be catching off this database connection, not just listing them but knowing even the concept that I should consider those error conditions? What could possibly go wrong with simple string concatenation when putting together a database query? Is there anything wrong with trusting the user to enter a quantity at a self checkout stand, and is there any input bounds checking to be done? (Like not permitting negative number quantities of produce, or not permitting the user to enter thirty different items in a row claiming them all to be cheap Idaho baking potatoes)
"Plz write me a hello_world.py" is pretty easy for an AI, but actually writing usable code seems quite challenging.
I am so old that I predate the internet going back to 1981 and I remember coding without google and stack overflow and similar resources; life was different back then and if you couldn't google how to invert a binary tree either you figured it out yourself (like a bad interview) or you researched paper textbooks or you didn't do it. Likewise AI will be a similar step, nobody will ever again type in by hand something for example like connect to a MSSQL DB in C#. But much as being able to google the algo to invert a binary tree didn't put all programmers out of work, it seems likely that not having to hand type simple stuff will put many people out of work.
I see the AI situation as very similar to the claim that photocopiers and NOLO publications will put all the lawyers out of business. Sure, some very small businessmen will get away with hand filing their own homemade incorporation papers without lawyers, and there will be a small amount of wage pressure, but overall they're not going to get wiped out.
Ask AI if "I" should use AWS, GCP, Azure, private cloud, or ? Instead of an answer explaining what "I" should do you can only get a generic groupthink comparison of "well, Azure is pricey for DB compared to AWS" and no interpretation.
However, the problem with being your own lawyer (or software developer) is the AI will never exceed its previous accomplishment but your goals and desires and requirements certainly will. So its a rising tide lifting all the boats situation. The game will be upped; you're not getting ahead by using AI you're just not going out of business, and the jobs are going to be even harder and higher paying. The cousin of this problem is if you have no idea what you're doing and randomly file legal papers in a best effort sense, just blindly trusting a liability-free book that may be out of date already, you'll eventually paint yourself into a corner if you try to set up your own LLC, c-corp, trust, will, etc, and as per above the AI was already operating at its limit so it won't be able to help and you're past your limit which is why you used the AI in the first place so you won't be able to help yourself or sometimes even understand the problem. Which is why human lawyers still practice...
In summary point 7 is I expect a lot of money will be made by humans cleaning up after AI "accidents". The purpose of the tech to "do stuff" beyond your skill level, which always turns into a disaster sooner or later.
A pretty good analogy from the pre vs post internet era is being able to blindly cut and paste code, perhaps an algo or an API, doesn't mean you can actually apply it correctly, use it, understand it, or troubleshoot it. And so it will be with AI, its just "nicer looking" code to cut and paste. But everyone who's ever self taught themselves an algo or an API knows they didn't really know anything when they cut and paste, the real learning came after.
"There's no such thing as human intelligence. We're just a bunch of organic machines following pre-programmed genetic codes and learned responses, with zero understanding of the true nature of our existence."
Person: I need a website to sell shoes. Here's my brand and logo. Here's a selection of what the product looks like.
5 seconds later.. AI: I've generated 5 prototype sites and deployed them to URL1, URL2 .. URL5. Please review and select one.
Person: URL3 looks great. Let's get that deployed and add all the shoes in this CSV. Monitor the site and make any changes to increase engagement and sales.
Now you might say: but what's so special about a shoe store? Payment, inventory, shopping carts, databases - it can all be the same as every other shoe store!
But that's the key takeaway here. If you work in something where the next deployment has much in common with the last one: then your job is at risk. But these jobs have been eaten by SaaS for a long time. An "online store as a service" was a thing long before AI.
Could AI help take over this type of software development even more? Sure. But that's like machines taking over dull industrial jobs. We shouldn't worry about being out of a job, we should be happy that we can do more interesting things than what the machines can do, just like we could stop weaving fabric.
Or maybe that's not realistic, and it will look somehow different. In any case, I think it will change, and you should be prepared to adapt to it.
If you ignore the whole development, your job might indeed be at risk, or just be at much lower value.
The first is whether the current generation of AI is capable of writing finished software based on an end user's natural language specification. I think the answer to that is no. There are too many loose ends and these AIs are not good enough at logical inference and root cause analysis, especially when the logic is extralingual.
The second question is whether these AIs can make developers more productive so that fewer developers are going to find employment. I'm not sure. Yes they will make us more productive, but the demand for developers may well outstrip the added supply. If something gets cheaper, people tend to use more of it.
But AI is slowly taking over these tasks, pretty much the same way the conveyor belt revolutionized our industry.
It is difficult to predict how fast this will happen, but somewhere within a 5-10 year period high level programming will definitively be different from what we are used to today.
As the AI technology improves, programming will be a matter of dictating what you want, from what sources and how it should be displayed.
Code will become better, stable, accessible for all and make us all high level programmers.
There will still be programmers, however they will most likely be much more specialized than what they are today.
LLMs are trained on existing contains which are presumably original so far. This means LLM can not think beyond what it has learnt from past resources.
- Would it be safe to assume LLM can not produce new original work?
- I worry internet will be soon will be flood with LLMs generated content which will again feed in to LLMs to train them further. IMO such feedback will hardly help, in fact that may make LLM more rigid in answers.
I know there is parallel effort going to detect machine generated content so that they dont rank on top but its is still in nascent phase.
Original code will be harder to duplicate. We can assume that those coders will use an LLM as a tool to speed up their development, but won't be able to trust it for mission critical items. Although, writing this I realized that an LLM may become an excellent way to bug test.
I guarantee that the C-Suite people are already salivating over the thought of firing all their programmers.
For people like me; probably not, but then, my skills don't seem to be in demand, so much.
Does the AI know your database?
Does the AI know your design, device, legal and other constraints?
No, and won't for any foreseeable future.
I worry about an AI that will LEARN your product and LEARN your database and LEARN your tribal knowledge. As a similie, I fear we are like a horse, in the era of Henry T. Ford.
On the other hand. Someone still needs to interact with ChatGPT or the likes. Someone needs to keep an eye on quality control. Someone needs to ASK chatgpt TO write the code. Its like programming on yet an even higher level. I just worry that, with that increased efficiency there wont be troves of teams writing code like today.
On the foot, chatgpt doesnt write perfect code and can introduce complex bugs that only the best of the best will be able to decipher, so there is that.
On the other foot, when you look at what joomla,wordpress,squarespace did to website design it is a very similar trajectory. There will always be niches
No. They have no idea what my codebase it, how to translate concepts like "audio is limited to users on free tier" into code etc.
> I worry about an AI that will LEARN your product and LEARN your database and LEARN your tribal knowledge.
It won't, not for any foreseeable future. It fails consistently even at the most simple of tasks. If you're afraid that you will be made relevant because we will be able to teach it those simple tasks?
> when you look at what joomla,wordpress,squarespace did to website design it is a very similar trajectory.
What did they do? Did they make you obsolete?
You wouldn't need project or product managers or designers. Heck, at that point, you don't need accountants either. Maybe you have a team that supervises the AI - but, really, you could have an AI do that. So you have the business owner(s) and zero employees. No more tech companies as everything can be done by AI, including CEO work.
What I actually expect: better tooling. We were not replaced when we went from punch cards to C - it just increased demand for programmers. AI may be the next tooling, but I only see more opportunity: solutions happening faster.
Not really. The answers for my questions are as far away now as they were 15 years ago.
None?
Search the net to find how many companies are cutting down their programmers' workforce and replacing it with AI.
Still none?
Search for anecdotal evidence on forums to find any programmer's job been replaced by AI.
All talks point to promises in the "future" only?
Rest assured. It happens when it happens.
Even if that happens maybe the ones that know how programming languages work will be the good ones, as today a good programmer knows about OS’, architecture or compilers.
There will be more people using the ai to program computers, but the ai interface may be so easy that it's no longer a skill.
I am trying to build my own Language Model using Pytorch and it is much easier than a lot of the stuff I've had to do as a traditional SWE over the years. My understanding is that the difficulty comes from training massive models that cannot fit on a single GPU efficiently, so more of a software problem than a maths problem.
Am I suffering from the Dunning-Kruger effect ? Or is the genius of the field in the combination of this limited set of techniques ?