Im aware of several AI assistance tools and using LLMs to generate (hopefully) runnable code but has anyone used the available tools to significant impact? I mean can I delegate a sprint ticket to an AI yet? I can see us getting there. A suitably detailed feature ticket isn't much more than a prompt. The ticket would need to get down to a code level in specification of what needed to be done. Its possible on paper sure.
Fun question: AI generates some code and its deployed to production. Theres a bug found in the logs and a dev sits down to analyze the fix. Does the dev correct the prompt that generated the code and commit that prompt, or does the dev make a code fix and commit that?
Right, training the models will still be a thing, for libraries, systems, service semantics, etc. Like with robot tenders, a smaller number, but still there.
I find one of them to be pretty good at autocompleting for-loops and simple statements for me. Because it's driven from the same tab key that I use to format code with, I also spend time dismissing suggestions with the escape key. As a power tool it's OK.
| A suitably detailed feature ticket isn't much more than a prompt. The ticket would need to get down to a code level in specification of what needed to be done.
This is the part I think folks underestimate when saying that "Coding as a career might be over". Gen AI can write decent code, but constructing software from code, and coming up with implementation ideas and turning those into tickets still feels far away at least from my experience using ChatGPT and CoPilot. Also those systems were trained on code but I don't know that they have the best idea of quality of the code they were trained on.
In my experience and at this point in my career ark I can say that "coding" itself hasn't been the hardest part of the job in ... 20 years? Coding was never difficult. Understanding the rats nest of code that comprises a product is a big barrier. The whole "I know this looks batshit crazy but we do it this way for a business reason" and it goes on and on. Not to mention the database full of tables that are gigantic and nobody knows what they're for anymore (This company maybe your cell carrier I kid you not).
In all seriousness though, an AI at this point can do some cool things with some handholding, prompt refinement, etc such that its almost a Jr. Dev. Both more or less went to school for CS and both are minimally experienced but have a lot of code to look at and draw conclusions from. Neither know anything about the business domain. Both generate candidate solutions that will at best require some additional guidance.
Personally, I don't see things getting too far beyond this in the near term. Perhaps AI becomes a very good Jr Dev who can handle "write me a function with inputs Foo,Bar that returns the [biz logic] as Baz" and the "coders" modern job is to generate maybe 50% of the necessary code where the other 50% is the integration.
Who knows ... but I dont see people saying the things I think people would say if anyone was actually doing this for realsies.
Coding well is very difficult, that is one of the reasons why all of our software is still unreliable and ridden with bugs. I agree its not the hardest problem in product development.
I hope he's right. I for one would be willing to sacrifice my career and find another one if it means everyone gets unlimited dirt-cheap ultra-high-quality software -- such an offering would be great for the world!
Modern "AI" is pretty cool at what it does, but I do think we're going to hit a bit of a slow down in the not too distant future. Everyone will adapt to using it as the tool it is, but I wouldn't let it dissuade you from pursuing development. There is still plenty of need for developers. Play around and find your Niche!
It is becoming more and more difficult to find a role in IT that won't be affected by AI in the long run. I'm sure there are some out there, just difficult to see them with clarity. There will always be a need for some folks but not as many.
AI will need to be managed/implemented? Perhaps become an AI Manager of some kind helping deploy it into industry. I'm sure it will need to be cared for, hardware breaks, power outages, tuning, environmental controls, etc. There is a whole industry that will need to support it until Robots/Androids can do that job.
It could be a rough transition. But on the bright side, once AGIs get capable enough, maybe they'll leave jobs to us that they find too boring to work on...
This is certainly overhyped or even hyperbolic to claim, unless the real claim is that AI will maintain its own source code, and presumably also improve on it, which is a massive claim at the moment.
Yes, of course, but I would still like someone to make him clarify if he means every single programming job or changing the landscape because we'll still need AI engineers. If it's the first one then the claim is massive and I think would require some evidence.
What he actually said (according to TFA anyway, I haven't verified he actually said this):
"It is our job to create computing technology such that nobody has to program. And that the programming language is human, everybody in the world is now a programmer. This is the miracle of artificial intelligence."
How do we get from this to "future of coding as a career is dead"?
I'm mildly surprised Jensen Huang hasn't either been fired or isn't facing jail time after everything that has happened.
Between...
* Trying to falsely sell an AI future that won't come in his lifetime (hes in his 60s), and his company almost has nothing to do with this future; GPUs are a horrible stopgap for purpose-built hardware, which companies already sell and have been deployed in prod for years.
* Selling GPUs that literally are fire hazards and then helped lead the effort to hide this and threaten anyone willing to cover it in the tech news; this only lead to Streisanding it hard.
* Covering up the whole Samsung fab failure where he blamed the lack of cards on retail shelves was because of "Bitcoin miners scalping them", and then opened up him and his company up to litigation over this by their own shareholders, and then they lost in court and were forced to admit to shareholders that the cards had never existed in the first place; this one is extra insane, because a sale is a sale (doesn't matter who the cards got sold to once they leave Nvidia partner hands), but Bitcoin miners would have mined with them and Bitcoin mining on GPUs had already been dead for several years by then and Nvidia GPUs have horrible performance when it comes to such tasks; scalping or mining would have been a less profitable proposition for miners, and he also slandered miners by lumping them in with scalpers with such a flippant remark.
* Fought the Mellanox acquisition internally until someone sat him down and explained to him that everyone had a high bandwidth low latency network except Nvidia and that AMD had used Mellanox for external Hypertransport for years (but transitioned to external PCI-E as part of the Infinity Fabric rebranding) and a lot of smaller supercomputer-esque deployments switched to generic RDMA over Ethernet instead of Mellanox Infiniband, and if they want to survive at all in the enterprise space, they need this, and due to everyone switching, Mellanox was now a cheap and effective acquisition. He then relented, purchased Mellanox, and started claiming it was his idea the whole time.
... I'm still not sure why Nvidia continues to keep him around. He only owns about 4% of the company, and only about 15% of the shares are part of the Jensen personality cult. Ditching him and hiring someone actually on board with the Nvidia mission would be a huge boon to the company.
I think we might be seeing the end of hyper specialisation. The need for pure coders will decrease in the same way pure scrum masters, business analysts or project managers did. The skills are still there and used but they are aspects of roles rather than the whole role.
That is my optimistic read on AI. Return of the generalists (note - don't read that as a 'jack of all trades' but rather someone who is very good at 2-5 things rather than an expert at 1-2). Have a broader base and know how to use AI tools to do the prescriptive parts.
It's an incomplete answer that's partially true and will become more true over time.
Software development will be effectively dead when enough systems can adequately code, define, and run themselves with minimal human design or operations attention. This is one of key (gradual) thresholds of the technological singularity.
To accomplish this will require far more than just code completion, but AI-operated requirements capture, architecture, feature prioritization, configuration management, operations management, capacity planning, debugging, and optimization.
Further still off in the future (perhaps 20 years), it will gradually eat down the stack to include algorithmically-designed programming languages and representations of languages, operating systems, and the ASIC silicon pipeline that run the things.
The pace of technology advancement is about to exceed human control and human understanding, and it also means that most people will be unemployed or under-employed with really shit jobs and very poor, while there will be a few trillionaires.
The alligator leather jacket he's wearing is absolute fire. In fact, I don't know much about Jensen, but his leather jacket game is on point. Are there any other CEOs who go as hard with leather jackets as he does?
Yeah but dragging the requirements out of the user and then managing their unrealistic expectations is something I can't really see AI doing in any of its current or likely forms.
We can all see how effective AI coding assistants are, they are very good.
They save a massive ammount of typing, can convert between different programming languages and even provide reasonable advice on how to structure and organize code, its not a mystery when it is sitting there in your IDE all day.
Do they replace programmers, I doubt it, i think they only accelerate programmers, this level of assistance will open programming up to a much wider pool of people, great time to learn to code.
I am not sure but the CEO of nvidia might be slightly biased, and possibly out of touch. Not saying it’s not possible, but this is the last person I would ask.
I think everyone is possibly missing something here, that perhaps this is not about AI's generating code and testing and debugging it, it's about the AI _becoming_ the app. You don't write a program, you just tell the AI what you want it to do and it does it. For example, "Hey AI, I'm going out for a while, please keep track of the speed and distance and watt-hours while I'm riding my ebike and give me a summary of the total distance and average speed for the times I'm riding the bike and do the same for the times I'm off the bike and walking." You could write an app to do that, you could tell the AI to write an app to do that, but what if you could just tell the AI to do it and it does it, no app needed? It figures out how to access the GPS on your phone, it figures out how to use the accelerometer to detect whether you are walking or riding, it figures out how to get the ebike to send power data, and it figures out how to create the summary data, no code needed. Initially this may only be possible for simple applications like the one I just described, but eventually it would probably be possible with much more complex applications like taking in business data and producing reports and recommending changes.
As Tim Cook famously said, why don't you buy your mom an iPhone and tell her to tell it "Hey AI, I'm going out for a while, please keep track of the speed and distance and watt-hours while I'm riding my e-bike and give me a summary of the total distance and average speed for the times I'm riding the bike and do the same for the times I'm off the bike and walking." It just works.
36 comments
[ 304 ms ] story [ 2361 ms ] threadFun question: AI generates some code and its deployed to production. Theres a bug found in the logs and a dev sits down to analyze the fix. Does the dev correct the prompt that generated the code and commit that prompt, or does the dev make a code fix and commit that?
I find one of them to be pretty good at autocompleting for-loops and simple statements for me. Because it's driven from the same tab key that I use to format code with, I also spend time dismissing suggestions with the escape key. As a power tool it's OK.
This is the part I think folks underestimate when saying that "Coding as a career might be over". Gen AI can write decent code, but constructing software from code, and coming up with implementation ideas and turning those into tickets still feels far away at least from my experience using ChatGPT and CoPilot. Also those systems were trained on code but I don't know that they have the best idea of quality of the code they were trained on.
In all seriousness though, an AI at this point can do some cool things with some handholding, prompt refinement, etc such that its almost a Jr. Dev. Both more or less went to school for CS and both are minimally experienced but have a lot of code to look at and draw conclusions from. Neither know anything about the business domain. Both generate candidate solutions that will at best require some additional guidance.
Personally, I don't see things getting too far beyond this in the near term. Perhaps AI becomes a very good Jr Dev who can handle "write me a function with inputs Foo,Bar that returns the [biz logic] as Baz" and the "coders" modern job is to generate maybe 50% of the necessary code where the other 50% is the integration.
Who knows ... but I dont see people saying the things I think people would say if anyone was actually doing this for realsies.
But also... what else could I pivot to instead that doesn't also have the same problem?
AI will need to be managed/implemented? Perhaps become an AI Manager of some kind helping deploy it into industry. I'm sure it will need to be cared for, hardware breaks, power outages, tuning, environmental controls, etc. There is a whole industry that will need to support it until Robots/Androids can do that job.
I'm semi retired and I am worried for you too.
Here’s an example, companies still hire technical writers. Even though there are so many AI text generators out there.
"It is our job to create computing technology such that nobody has to program. And that the programming language is human, everybody in the world is now a programmer. This is the miracle of artificial intelligence."
How do we get from this to "future of coding as a career is dead"?
Between...
* Trying to falsely sell an AI future that won't come in his lifetime (hes in his 60s), and his company almost has nothing to do with this future; GPUs are a horrible stopgap for purpose-built hardware, which companies already sell and have been deployed in prod for years.
* Selling GPUs that literally are fire hazards and then helped lead the effort to hide this and threaten anyone willing to cover it in the tech news; this only lead to Streisanding it hard.
* Covering up the whole Samsung fab failure where he blamed the lack of cards on retail shelves was because of "Bitcoin miners scalping them", and then opened up him and his company up to litigation over this by their own shareholders, and then they lost in court and were forced to admit to shareholders that the cards had never existed in the first place; this one is extra insane, because a sale is a sale (doesn't matter who the cards got sold to once they leave Nvidia partner hands), but Bitcoin miners would have mined with them and Bitcoin mining on GPUs had already been dead for several years by then and Nvidia GPUs have horrible performance when it comes to such tasks; scalping or mining would have been a less profitable proposition for miners, and he also slandered miners by lumping them in with scalpers with such a flippant remark.
* Fought the Mellanox acquisition internally until someone sat him down and explained to him that everyone had a high bandwidth low latency network except Nvidia and that AMD had used Mellanox for external Hypertransport for years (but transitioned to external PCI-E as part of the Infinity Fabric rebranding) and a lot of smaller supercomputer-esque deployments switched to generic RDMA over Ethernet instead of Mellanox Infiniband, and if they want to survive at all in the enterprise space, they need this, and due to everyone switching, Mellanox was now a cheap and effective acquisition. He then relented, purchased Mellanox, and started claiming it was his idea the whole time.
... I'm still not sure why Nvidia continues to keep him around. He only owns about 4% of the company, and only about 15% of the shares are part of the Jensen personality cult. Ditching him and hiring someone actually on board with the Nvidia mission would be a huge boon to the company.
That is my optimistic read on AI. Return of the generalists (note - don't read that as a 'jack of all trades' but rather someone who is very good at 2-5 things rather than an expert at 1-2). Have a broader base and know how to use AI tools to do the prescriptive parts.
Software development will be effectively dead when enough systems can adequately code, define, and run themselves with minimal human design or operations attention. This is one of key (gradual) thresholds of the technological singularity.
To accomplish this will require far more than just code completion, but AI-operated requirements capture, architecture, feature prioritization, configuration management, operations management, capacity planning, debugging, and optimization.
Further still off in the future (perhaps 20 years), it will gradually eat down the stack to include algorithmically-designed programming languages and representations of languages, operating systems, and the ASIC silicon pipeline that run the things.
The pace of technology advancement is about to exceed human control and human understanding, and it also means that most people will be unemployed or under-employed with really shit jobs and very poor, while there will be a few trillionaires.