AI & Developers - Will We Survive?

00:00:00: - So, 2023 is over,

00:00:02: and it of course has been the year of AI,

00:00:06: even though it, of course, already all started in 2022,

00:00:11: when ChatGPT was released in November '22.

00:00:16: But the AI pace really picked up and accelerated in '23,

00:00:22: when loads of new AI services and products were released

00:00:28: when barely a week went by

00:00:31: without Microsoft announcing some new Copilot,

00:00:35: and therefore, of course, AI is here,

00:00:39: and AI is here to stay and will very, very likely

00:00:43: only become more and more powerful and hopefully helpful.

00:00:48: But speaking of that helpfulness,

00:00:51: what does AI mean for us developers?

00:00:55: Will it replace us?

00:00:57: Will it replace us this year, the next year,

00:01:00: or maybe not at all?

00:01:02: That's exactly what I wanna share my opinion on.

00:01:07: I see three main ways of AI replacing developers

00:01:13: or having an impact on developers, you could say.

00:01:17: And the first way it will have an impact on developers

00:01:22: is mainly as a tool that as developers,

00:01:27: we can use AI services like ChatGPT,

00:01:31: but of course also GitHub Copilot, which under the hood,

00:01:35: uses the open AI APIs.

00:01:39: So it uses ChatGPT under the hood, you could say.

00:01:43: It uses the same APIs

00:01:45: and other tools like Midjourney.

00:01:47: As developers, we can use them to speed up our workflow

00:01:52: to become more efficient,

00:01:55: because AI writes parts of our code,

00:01:58: writes code blocks,

00:02:00: gives us suggestions,

00:02:02: helps us debug errors,

00:02:04: or if we're thinking about tools like Midjourney,

00:02:08: helps us generate assets,

00:02:11: which we might need to build prototypes

00:02:13: and things like that.

00:02:15: And that's therefore one way

00:02:17: AI could have an impact on developers,

00:02:21: and I'd argue, already has an impact on developers

00:02:26: because that is pretty much the the present,

00:02:29: it's the current state that we have such AI tools

00:02:34: that do make us developers more efficient.

00:02:38: So that's one possible way of AI impacting developers.

00:02:43: Maybe we stay in that world

00:02:45: and that's all we can expect from that AI

00:02:49: or revolution as developers.

00:02:51: And I will get back to that later.

00:02:53: But of course, there also are more extreme scenarios,

00:02:57: more dangerous scenarios, you could say.

00:03:00: Well, for example,

00:03:02: the second main way of AI having an impact

00:03:07: on developers I can think of

00:03:09: is that AI writes much more code than it does today.

00:03:15: Because today, as I just explained, it's mostly a utility,

00:03:20: a helpful tool that generates some code,

00:03:23: but not all the code.

00:03:25: And in the future, maybe even in a not too distant future,

00:03:30: AI may be able to generate entire applications

00:03:36: so that it writes all the code of an application,

00:03:39: not just individual snippets or code blocks.

00:03:42: And therefore, of course, it might be able to get rid of

00:03:47: a lot more developers,

00:03:48: almost all developers, you could say,

00:03:51: because we maybe just need a few that tell the AI what to do

00:03:55: and supervise the results.

00:03:58: But we might not need as many developers as we have today

00:04:01: because AI writes most of the code.

00:04:04: So that's another scenario

00:04:05: which could be dangerous for us developers.

00:04:08: And I will of course also get back to that.

00:04:11: But first, I wanna outline

00:04:12: the third scenario I can think of,

00:04:15: and that would be a world where we maybe don't need

00:04:19: that much code to be written.

00:04:22: Because why would we assume that the future

00:04:27: is about AI writing code?

00:04:30: Why is it not about AI applications

00:04:34: that can then directly do certain things?

00:04:37: So we don't need specialized applications

00:04:40: to perform certain tasks.

00:04:42: Instead, we might have a couple of AI applications

00:04:46: with which we interact

00:04:48: through voice input or written prompts,

00:04:51: which are then able to do all kinds of things.

00:04:54: So instead of a hotel booking website,

00:04:56: and a flight booking website,

00:04:58: and a website for doing online groceries,

00:05:02: we might have one AI service,

00:05:05: which of course would be an application,

00:05:07: but it would just be one,

00:05:08: with which we interact,

00:05:10: which we then tell that we wanna book a vacation,

00:05:13: that we need a hotel, some flights,

00:05:15: and that we also want to order

00:05:17: some milk and eggs for tomorrow.

00:05:19: And that of course would also be a possible future,

00:05:23: and it would be a future in which we need,

00:05:26: basically, the developers that work

00:05:28: on those AI applications and maintain them,

00:05:32: even that might be doable by some AI.

00:05:35: And hence, the entire idea of people writing code

00:05:40: and building applications

00:05:42: might be an idea from the past in that future.

00:05:47: But how likely are these different futures?

00:05:52: How likely is it that we enter one of these three scenarios?

00:05:57: And let's maybe start with the most extreme scenario.

00:06:01: The third one, where we don't really

00:06:04: have a lot of applications,

00:06:06: but where we instead have those AI applications,

00:06:10: those AI services which perform the majority of work

00:06:15: or which do the majority of work,

00:06:17: and where we therefore need basically no developers.

00:06:21: How likely is that?

00:06:23: And I personally don't think it's too likely.

00:06:27: Definitely not in the very near future,

00:06:30: but maybe also never.

00:06:33: Maybe this is never the future we'll live in

00:06:36: because, of course,

00:06:38: AI will become more capable and powerful,

00:06:41: and we will have versatile AI applications

00:06:45: with which we can interact.

00:06:48: The question just is if all interactions are best done

00:06:54: with such an AI application,

00:06:56: and if we want such AI applications to do everything,

00:07:01: do you want an AI you talk to, to check your account balance

00:07:06: instead of logging into your online banking account

00:07:10: and looking at it there?

00:07:11: Do you want an AI to have access to your bank account?

00:07:15: Is the question, I guess.

00:07:18: Do you really want to book a vacation

00:07:22: by telling an AI what you wanna do?

00:07:24: Is that really how you wanna do it?

00:07:26: At least for me, it's not.

00:07:27: I like the process of picking a hotel,

00:07:31: looking for different flights,

00:07:33: reading about the destination I'm traveling to.

00:07:37: That, for me, all belongs to that vacation experience.

00:07:41: And it's not just the experience.

00:07:43: I don't wanna rely on an AI picking the best flight

00:07:48: and the best hotel for me.

00:07:50: I wanna choose it myself.

00:07:52: And I'm fully aware that of course,

00:07:55: on those hotel booking websites, we also get influenced.

00:07:59: The way those hotel recommendations are ranked

00:08:01: is not necessarily the best way for us,

00:08:05: but at least we have a choice.

00:08:06: We can go through them.

00:08:08: I don't wanna rely on an AI to do that for me.

00:08:12: I don't trust AIs that much. I can't look into it.

00:08:15: I don't know why it picked certain options.

00:08:19: And these, of course, are very specific examples,

00:08:22: but you could come up with similar examples

00:08:25: for basically all industries.

00:08:28: And then it's also also possible

00:08:31: that we have certain application areas, certain industries,

00:08:36: where such a versatile AI

00:08:38: might not even be the best solution.

00:08:41: It might in some cases,

00:08:43: be too complex to always interact with an AI

00:08:47: instead of a specialized application

00:08:50: that does a specific job well.

00:08:52: So chances are high that

00:08:54: in probably a lot of scenarios and use cases,

00:08:58: such a versatile AI application

00:09:01: would be less reliable and less efficient

00:09:05: than a specialized application that does one thing

00:09:09: and does that thing well.

00:09:11: For example, I don't want an AI application

00:09:14: which I have to tell to scan the document

00:09:17: I just put into my scanner.

00:09:19: Having a dedicated application for that

00:09:21: makes a lot of sense to me

00:09:23: because I don't wanna order groceries

00:09:26: from that same application.

00:09:27: I just want to scan the document I put into my scanner.

00:09:31: And again, it's just a made up example,

00:09:33: but you probably get my point.

00:09:36: And that's even not all that could be mentioned here.

00:09:39: This is maybe just my personal preference

00:09:42: and others might disagree.

00:09:44: But what about accountability and responsibility?

00:09:48: If we rely on versatile AI applications,

00:09:52: it basically means that we trust them to do things right.

00:09:57: Because again, looking into them is pretty hard.

00:10:00: Sure, we might be able to come up with ways of logging

00:10:05: how an AI came to the conclusion it came to,

00:10:08: but even then you have to trust it

00:10:10: that this logging is done right,

00:10:12: because there is no deterministic code that leads to that.

00:10:16: Instead, we all know

00:10:17: that if we give an AI the same prompt twice,

00:10:21: we might not necessarily get the same result twice.

00:10:25: But even if that would change,

00:10:27: we still would have an AI doing things,

00:10:30: and an AI is probably not accountable for its actions.

00:10:35: And that's a big difference compared to a human.

00:10:38: If you got humans writing code for an application,

00:10:41: you have responsibility,

00:10:43: you have accountability.

00:10:45: And that of course is a major factor

00:10:48: for so many industries.

00:10:51: And kind of related to that,

00:10:53: it of course is probably fair to assume

00:10:56: that even if we would have a perfect world

00:10:59: where we got a super powerful AI

00:11:02: that is accountable, that we all wanna use,

00:11:05: even if that would be the world we'd be living in,

00:11:08: we'd still have a lot of industries,

00:11:11: and government services, and so on,

00:11:15: that would probably not use it in the short or even midterm

00:11:20: because we all know, at least here in Germany,

00:11:23: but I guess it's the same all over the world,

00:11:26: how quick certain industries,

00:11:29: and especially government services have been

00:11:32: to adopt digitalization and the internet.

00:11:36: Here in Germany, there still is a lot of stuff

00:11:40: that has to be done on paper, not over the internet,

00:11:43: not using digitalization.

00:11:45: And we can only imagine how quick

00:11:48: AI services would be adopted there.

00:11:51: And of course then, we also would've other problems

00:11:54: like data privacy and so on.

00:11:56: So, I am very confident that in the near and midterm,

00:12:01: and maybe never,

00:12:02: we'll be living in a world

00:12:04: where we only interact with AI services.

00:12:09: But that was just one extreme world.

00:12:11: Another pretty extreme world for us developers,

00:12:15: would be one where AI writes

00:12:18: basically all the application code that must be written.

00:12:22: In such a world, we would still have some developers

00:12:25: that supervise those results,

00:12:27: and check them and maybe fix tiny errors,

00:12:30: or write very specific niche code,

00:12:34: but we wouldn't have as many developers as we do today.

00:12:38: That would be another possible scenario.

00:12:41: But here, again,

00:12:42: I'm doubtful if we'll be living

00:12:45: in such a world anytime soon.

00:12:47: For a couple of reasons.

00:12:49: Now, one reason is essentially the same

00:12:51: as I just stated before,

00:12:52: accountability, and then speed of adoption, and so on.

00:12:56: That would be the same

00:12:58: because if AI writes all of the code,

00:13:01: it again, would be the question,

00:13:03: who's responsible for that code?

00:13:05: Who's responsible if it contains critical bugs

00:13:08: that might lead to severe errors?

00:13:12: Now, you could say, |The people who supervise the AI

00:13:16: and whose responsibility is to check the code,"

00:13:18: but how many developers would you need for such a task

00:13:23: if we're dealing with a complex application?

00:13:25: Would you really gain that much by using an AI then,

00:13:29: or would it not be better to have

00:13:31: like a similar amount of developers who use AI support

00:13:36: but write the code themselves?

00:13:39: Because we all know that

00:13:41: diving into code we have not written,

00:13:44: of course, takes way longer to understand it

00:13:47: than if we write code ourselves.

00:13:50: It's a natural thing.

00:13:51: And therefore, this problem, at least partially,

00:13:55: might still exist.

00:13:56: But I have an even bigger doubt.

00:13:59: I am not convinced that anytime soon,

00:14:02: AI will be capable enough

00:14:05: to write entire applications on its own.

00:14:08: Sure, we have all seen those examples

00:14:11: where people drew a basic website onto a napkin,

00:14:16: scanned that, and then the AI built that.

00:14:18: But let's be honest,

00:14:20: these were super simple and pretty ugly examples.

00:14:23: This is not really impressive.

00:14:26: I mean, it's impressive

00:14:27: that we have technology that can do that,

00:14:29: but it's not something we have to fear

00:14:32: when it comes to that technology

00:14:34: replacing us for complex scenarios.

00:14:37: Because for basic websites like this,

00:14:40: you didn't really need developers before.

00:14:43: You could just use a page builder like Wix,

00:14:46: or some content management system,

00:14:49: and you could even build

00:14:50: way more complex websites with those.

00:14:53: But when we talk about more complex websites

00:14:56: or applications in general,

00:14:59: I am really not sure if we'll soon be in a world

00:15:03: where AI can write all that code,

00:15:06: make sure that all the different modules work together,

00:15:09: extend the code in an error-free way if needed,

00:15:14: and cover all the different niche, cases,

00:15:17: and specifications of those different applications.

00:15:22: Because of course, every application is different

00:15:25: and has certain conditions that must be met,

00:15:28: has certain specifications that must be implemented.

00:15:33: And AI doing all that in a pretty error-free way,

00:15:38: in a way that's then maintainable and extendable,

00:15:42: well, surprise me, but I'm really not convinced

00:15:46: that this is a world we'll be living in anytime soon.

00:15:50: I see AI as a tool that can help us,

00:15:53: that can write some code,

00:15:55: but writing entire applications in a reliable, good way,

00:16:00: I'm not so sure.

00:16:01: And that then leads me to the last option,

00:16:04: this first possible future I outlined before,

00:16:08: where AI is mainly a tool for developers,

00:16:12: where it writes some code but not all the code,

00:16:16: where we can use it to become more efficient.

00:16:19: And I think that's the future we'll have in the near future.

00:16:25: And it's actually already the present, you could say,

00:16:28: because of course we have GitHub Copilot and so on.

00:16:32: And for me, personally, it made me more efficient,

00:16:36: but not by getting rid of tasks I like to do,

00:16:41: but instead, by giving me smart suggestions,

00:16:44: by speeding up the process of writing boilerplate code,

00:16:48: and therefore by really allowing me

00:16:51: to get more done in less time.

00:16:53: And that's the kicker here.

00:16:55: Of course, you could now argue

00:16:57: that this will also replace developers.

00:17:00: Because if fewer developers can get the same work done

00:17:04: that previously was done by more developers,

00:17:08: there's room to decrease the number of developers, right?

00:17:13: Yeah, maybe that is what we'll see in the short term.

00:17:16: But in the mid and long term, chances are,

00:17:20: as it has always been in the history of humankind,

00:17:24: that we'll actually see more jobs

00:17:27: and more opportunities for people

00:17:29: because we have better tools.

00:17:32: In our entire history,

00:17:34: we always kept inventing better and better tools,

00:17:37: and that allowed people to become more efficient.

00:17:41: And maybe if one job then needed less workforce,

00:17:45: that same workforce

00:17:46: was able to either move to a different job

00:17:50: or use those tools to innovate more,

00:17:54: to produce extra growth.

00:17:57: And I think that is what we'll see here.

00:18:01: If developers can be more efficient,

00:18:04: we might have more industries and more companies

00:18:08: moving into the digital space and building applications.

00:18:12: Because a company that previously

00:18:14: might have not been able to afford

00:18:17: a developer staff of 20 developers,

00:18:20: might now be able to expand its business

00:18:25: by building applications that can now be built

00:18:28: with only 10 developers.

00:18:29: So it opens up new opportunities.

00:18:32: And of course, all the opens up opportunities

00:18:35: for individual developers.

00:18:37: Starting your own business,

00:18:39: your own software as a service business, for example,

00:18:42: is probably going to get a lot easier

00:18:45: because you can now get more work done on your own

00:18:48: and you don't need to hire a second or a third developer

00:18:53: to build your initial prototype.

00:18:55: So, I'm not even sure if it will lead

00:18:58: to a decrease in developer workforce.

00:19:01: Maybe in the short term.

00:19:03: But I personally don't think

00:19:05: that it will do that in the mid and long term.

00:19:08: Now, ultimately, of course, like everyone else,

00:19:11: no matter how convinced they may sound,

00:19:13: nobody can tell you with certainty

00:19:17: what the future will hold.

00:19:19: We can only take a look at the past and see that there,

00:19:23: we always had new tools and new technologies,

00:19:27: and they always change things,

00:19:29: but they also

00:19:31: pretty much always changed things to the better,

00:19:35: made us more efficient,

00:19:36: opened up more opportunities.

00:19:38: And therefore,

00:19:40: I'm very positive about the future as a developer.

00:19:44: And I think that AI can be an amazing tool

00:19:48: to accelerate us as a developer,

00:19:51: to make us more efficient,

00:19:52: and give us more opportunities that we can leverage.