Future of Software Engineering in the AI Era

[00:02] who works as a staff engineer in one of the prominent fintech companies here in US. They use Claude code for software development. They have a huge monolith with complex logic and thousands of lines of code. Yet, in last 1 month, 95%

[00:18] of pull requests that he has generated is completely written by AI. This is not AI assisted coding. This is AI first development where his role is to give development where his role is to give prompt intent and necessary context to

[00:32] Claude code and it will write the code the AI will write the code and all he will do is validate the output. When I asked him, "Do you find this new asked him, "Do you find this new paradigm of using AI for code writing is

[00:45] transformative?" He said, "No, it's not transformative. It is an absolute game-changer." While some may find this very powerful, if you are a software developer, this will sound very scary. You will be thinking, "Okay, what's

[00:58] going to happen to my job?" I also have the same question. We have this software and AI consultancy company called Ethic Technologies. We have a team of 50 people. We have built hundreds of software projects. And now

[01:13] we are evolving as a company as we work on software and AI projects for our clients. And this is the question that we have been asking to ourselves constantly. Based on my past industry experience also, I have certain views on

[01:27] how this evolution is going on. And I have many friends who are working in big tech companies who are industry leaders, etc. So, after having conversations with all those people, I have came up with some analysis which I'm going to share

[01:41] in this video. As a first step, let me show you a quick demo of Cursor so that we understand the reality of how good AI is in terms of writing code. So, here I have a sample blogging application. It has a back end written

[01:56] in fast API in Python. Front-end is a TypeScript code basically written in Next.js framework. And there is a database. So, this is a complete three-tier application. Unless I'm a developer working on this

[02:10] application, and I want to search for my blogs, I search it, and I see this error. Okay, so there is a bug, and I have been asked to fix this bug. Previously, I will debug this thing, you know, step through the code, and do my

[02:26] usual activity to find a fix. But now, in this world where you have AI as an assistant, what you can do is, especially when it comes to Cursor, is see, you're seeing this error, correct? So, you select this, you press Ctrl L,

[02:40] it will pull that error trace here. You can also take a screenshot of your error, okay? So, I will select this capture area screenshot, and I will select this, see. And notice what happened. It pulled that

[02:54] screen here. And I'll just say, "Fix this." So now, as a software engineer, my job is to provide a context to this AI. I said, "Okay, where I'm facing the error?" I gave a relevant screenshot,

[03:08] and it started working on it. Now, in Cursor, I have a Pro subscription, so I'm using the best model as of now, which is Opus 4.5. And apparently, see, it already found a fix. It found that there is some kind of

[03:22] parameter mismatch, okay? And it is working on fixing it. So now, I will click on review button. I will review the code change, and I will say, You can also click here, and if you click review agent changes, the other

[03:39] LLM will review the code change. So, the actual code change is made by LLM one. And when you say review agent changes, some different LLM, LLM two, is

[03:51] reviewing those code changes, okay? So, here now when I search for AI once you see? The issue is fixed. It's so easy.

[04:03] So, you can see how AI is making us productive. Now, let's say I want to implement a feature. So, I have this article and I want upvote button. Or let's say I want a functionality to add comments.

[04:16] So, once again, I can open a new chat window and I can use this DOM selector, So, I can say that, okay, below this element. So, it You see this

[04:28] div element? It picked that particular element. I'm saying below this add functionality to add comments.

[04:40] It gets even more powerful when you use your voice, okay? So, I have this your voice, okay? So, I have this Whisper Flow tool using which I can talk to it, okay? So, I will say, let me just enable that. Make sure to add both

[04:55] the title and description for the comment. So, when I leave that key, see? it. I'm just talking, you know, using the voice. So, you can use voice to give all these instructions to AI. And of course, this is a complete change. It's

[05:11] database, everything. So, it's going to take some time. See, it is creating to-do's also. So, how as a programmer you will plan things, right? Create comment model, create comment sections, create this, create that.

[05:25] It is just doing a proper planning and then implementing those changes. It took It wrote more than 500 lines of code. See, these are all the files that server, below the blog, see? I already see this

[05:41] comment. I gave very loose requirement. I did not give any mock-up for the UI, etc. It just used its average internet knowledge and implemented this feature, which looks pretty good. And I can just say, "Okay, nice article.

[05:56] Amazing insights." And just post. See? This is posted. So, I go back and when I click on it, the comments are persisting in the database. So, the database has this now

[06:11] uh table. You can see. All of this is done through a single prompt. I did not do any iteration. Now, this is very powerful, but at the same time it is scary. So, let's go over our

[06:24] analysis. Let's understand this economic concept of Jevons paradox. What it says is as technological improvements make a resource more efficient to use, overall consumption of that resource often increases, not decreases.

[06:40] This has happened when the fuel efficiency of car was improved. People thought that fuel consumption, overall consumption, will decrease. But what happened is because it became more efficient, more and more people started

[06:54] efficient, more and more people started using cars and the overall, the total consumption of fuel increased. Same thing is going to happen with software coding. We are not seeing any decline in terms of the code that needs to be

[07:09] terms of the code that needs to be written, okay? So, the world needs more and better code. There are many companies who are using old technologies, mainframes and all kind of old technologies, right? And

[07:21] still uses mainframe. Now, they have this multi-year migration projects going on. When I was at Bloomberg, we had this Oracle migration project, which was going on for many years. It was never ending. Because making these changes,

[07:36] implementing them, and then deploying it to your customer is actually a long and tough process. Now, with AI, it will help speed up this process, okay? So, one thing is very clear, the world needs better code, more code. As

[07:53] writing code becomes efficient and cheaper, many small to medium-scale organizations, which were slow on their digital transformation because of their cost concerns and because of the delays, they will be willing to accelerate their

[08:09] So, they will write, let's say, mobile app, they will upgrade their infrastructure, cloud infrastructure, they will update their on. And overall, experimentation will lead

[08:23] And overall, experimentation will lead to innovation. Many companies, they do AB testing to release the features. Now, it's it's not easy, actually. So, right now, only some of the sophisticated companies are able to

[08:37] do it. But, with this software code writing becoming cheaper, now even small companies will be willing to do AB testing for releasing new to do AB testing for releasing new features. So, overall, the

[08:50] activity of writing code is going to increase exponentially. Now, what is our role as a human? Because you might have this question that, "Okay, I agree that world will need more code, but what if all that code is written by AI?" And

[09:05] your concern is valid. A lot of this code is going to be written by AI, but we still need humans as an orchestrator. What I mean by that is in my current flow, or the flow that some of my friends are using, you are giving

[09:20] business requirements, legacy hacks, constraints, architectural boundaries, design suggestions, suggestions on coding style, etc., to LLM. So, your coding style, etc., to LLM. So, your role as a human is to provide intent and

[09:35] role as a human is to provide intent and context to LLM. LLM will produce the code. You will validate the output. And when I say output, it means you are validating the code that LLM wrote, as well as you are

[09:48] executing the code and seeing that okay, it is doing the thing that you expect it to do, right? We fix that bug, so we verify that the bug is actually fixed. And this kind of iteration will go on. So, essentially, you are an

[10:01] orchestrator, okay? So, you are here, you are orchestrating You can be using Cloud Code, Cursor, and some of these tools, right? Like AWS, Data Dog, GCP, and so on. Whatever you use for software development, and you will be

[10:18] implementing the features that the business needs. So, we are not in a time period where this entire thing can be automated completely, because we still need to talk to customers who are humans. We still need to talk to

[10:32] business managers who are humans and provide all the context, years of knowledge. And also, because LLM can hallucinate, you need to validate its output, right? Otherwise, you'll deploy a code in a production, which can be

[10:46] very disastrous for your company. So, in terms of evolution, when I was discussing this point with my friend, see, right now, in 2026, we have different roles in IT. We have software development engineer,

[11:00] we have product owner, and so on. As years go by, some of these roles are going to merge and consolidate. So, I'm seeing that SDE and product owner role will probably consolidate, okay? So, now you will have

[11:16] this new role, and for the lack of better term, let's call it builder. Okay? So, builder is a person where 50% they are doing coding, okay? Coding and technology.

[11:30] And then 50% or whatever percent, like 30, 40, 20, 80, whatever, uh they are doing the work of product owner. So, let's say if there is a progressive themselves uh in building these other skills

[11:45] because code writing has become efficient. Previously, you used to take 8 hours. Now, same code is written in 4 hours with the help of AI. What do you do in remaining 4 hours? Your company will have some expectation. I was

[11:58] talking to uh a friend of mine who is a founder of one of the prominent consulting company based in Gujarat in India. And that's what we are discussing, right? He told me that he wants his

[12:12] staff, right? His team to do more because writing code has become efficient. So, now imagine there is this builder person who is doing technical work, who is also good good in communication or also if they're not

[12:27] themselves. They're building soft skills. They are taking interest in domain. So, let's say if you're building projects in retail industry, you are gaining knowledge in retail industry. Maybe you are doing some certification.

[12:40] You are talking to your business, your customers, and gaining all of that knowledge. And let's say in this four or five years, okay? And this is imaginary five years, okay? And this is imaginary timeline. So, from 2026 to 2030,

[12:53] uh now when you go to any job portal, you will see this consolidated role anything. I don't know what they are going to call it. But in the "Okay, you need to do coding. You need to do this. You need to be good in terms

[13:10] to do this. You need to be good in terms of using AI tools. So, you should be uh AI generalist, okay? Who can use AI tools not to write just code, but to tools not to write just code, but to draft requirements, to draft uh scope of

[13:24] work document, and so on." And my friend who works as a staff engineer, he told me the same thing. This is the career trajectory that he's seeing for himself. All right, so just to conclude, there will always be humans who will play this

[13:37] role of orchestrator. Number of jobs may reduce little bit, but it will not reduce to an extent that we think they're going to reduce because more and more companies will build software going forward, okay? So let's

[13:53] say today, if you analyze the entire world, and let's say today total 1 world, and let's say today total 1 million software projects are active. million software projects are active. Okay? But with AI, uh as time goes by,

[14:07] active. And these are new companies, new startups experimenting fast, uh old companies doing their migration faster. So right now, let's say Bloomberg had this migration for Oracle.

[14:21] Now they might be doing Oracle migration and five other migrations at the same time. Right now, they're not able to do it because of the capacity constraint, and you know, they have the pipeline. Now they can do things in parallel.

[14:33] So when you have more software projects, okay? And let's say here with 1 million software projects, total number of programmers who are working on these projects is, let's say, 5 million, okay? So 5 million 5 million developers are

[14:49] working on this. Now because AI is making code writing faster, programmers, right? For 1 million projects, let's say, projects, let's say, you need only let's say you have team of

[15:04] five per project. Now you need team of only one person. One person. Okay, one person per project. So now these 5 million jobs are gone. That's what we'll think. But

[15:18] number of projects have increased, okay? And this is one of the hypothesis that I Because number of projects have increased, if there are 5 million projects, and let's say one project requires one person, then number of jobs

[15:32] are again 5 million. Or let's say 4.8. They reduce a little Or let's say 4.8. They reduce a little bit, but because of increase in demand, and because and see, the hypothesis has this assumption

[15:44] that there will be human in every project, or there will be a role of human as an orchestrator. Okay? Now, tomorrow you say, "Oh, robots will come, they will take over." See, folks, this that future is far. Okay? That future is

[15:59] that future is far. Okay? That future is far, far, far away. Okay? People talk about all those doomsday. Things don't happen that faster as well. Okay? Let's take example of self-driving cars. This article is from 2013.

[16:14] Elon Musk predicted that 90% of car driving is going to be self-driving, where there is no driver in the seat. So, he said that this will happen in 3 years. So, that is in 2016, right? The

[16:29] article is in 2013. That's when he said this. So, by 2016, we should see 90% of cars being driven on its own without any driver. We are in 2026 right now. 10 years later,

[16:43] and I don't think even 5% of car driving is fully automated. You see some self-driving cars in San Francisco, etc. I I agree. But if you look at the overall usage of car driving in US,

[16:57] car driving in US, it's nowhere closer to closer to 90% even after 10 years of him predicting it. So, these things evolve slowly than what, you know, these leaders expect it to be.

[17:11] And leaders are known to create hype. Okay? So, we have to kind of accept this truth that the role of humans in software code development is going to software code development is going to stay for longer time period. That role

[17:24] is already evolving. So, it's not going to be same as what it used to be, okay? to be same as what it used to be, okay? It is evolving from just a code writer to a builder, but the role is still there, okay? The role of builder as a

[17:37] human orchestrator who can work with AI tools is going to remain. So, if you're worried about future, I think the next thing you can try to do is build soft skills, you know, upgrade yourself from software developer to builder who can

[17:51] work with AI tools, who can work with business, who takes deeper understanding finance, whatever sector you are interested in. Take deeper understanding interested in. Take deeper understanding in domain and become good at uh giving

[18:05] intent to AI and then getting the expected output from AI, validating it, deploying it, and so on. It is not just that the role of software developer and product owner is merging. There are many other roles. For example, AI engineer

[18:20] role. Nowadays, when software developers are building applications, they will inevitably uh include some AI features in it into it. So, they need to know this AI integration skills and so on. And we have this AI engineer boot camp

[18:36] which is mostly self-paced videos and many of our uh students have requested us to provide a live version of it or a

[18:48] uh cohort version where is there is more live interaction, etc. So, I'm conducting a little survey. It will be a huge favor, folks, if you can check video description and fill out a form. Once again, this form is to assess the

[19:02] demand of live AI boot camp on CodeBasics website and I want to know what are people's expectations. So, it will be a huge favor if you can fill out that form quickly. It will take only few minutes for you. Now, when does using

[19:17] LLM help the most? So, I was watching Martin Fowler's video on the future of software engineering in AI era, and I saw this interesting comment. There is a person who said that they got a legacy project

[19:30] who said that they got a legacy project and they did not know the logic. You where you join the company, you are given a project, and when you say, "Okay, where is the documentation? Or who is the person who worked on it?" And

[19:43] all you get as a response is silence. Okay? And then you have to kind of dig because the person who built that code left, and they have no documentation. So, this person had that kind of project, legacy project, a technology

[19:57] where nobody wants to work on it, and LLM worked wonders for them. We shipped new features on top of outdated code. So, if you're migrating to a new stack stack, let's say you have old code in C++ plus, now you're migrating it to

[20:11] Python. For that kind of use case, LLM works wonders. If you're working on who wrote that code, where is the documentation, what is the logic, what is the purpose of this hack, etc. LLM can give you reasonable assumption.

[20:26] And if you want to build fast prototypes, LLM help the most. LLM also prototypes, LLM help the most. LLM also helps when you have huge code base with complex logic. Yes, it helps, definitely. But these are the scenarios

[20:38] where using LLM is a no-brainer. That's it, folks. Let me know your take on this it, folks. Let me know your take on this analysis down in comments. I'm ready to discuss just in case if you have any questions or any different point of

[20:50] view. To summarize, the future of software development is not as scary as it has been portrayed in media. Definitely, roles are evolving, so people who are not changing or upgrading to new ways of working are going to lose

[21:05] to new ways of working are going to lose their job, but people who are having this progressive mindset and who are evolving to the needs of new time, where they're ready to contribute beyond just core writing, are definitely going to

[21:19] thrive. Thank you very much for watching this video.