What AI Can and Can’t Build on Its Own | James Edgell on TechUnhinged

00:00 Welcome back to another episode of TechUnhinged where tech gets human powered by code district. I'm your host Rabia Javeed and my guest today is James Edgell, a

00:06 technology executive and founder with more than 25 years of experience building and scaling software from IBM

00:12 Motorola in McDonald's to Deloitte and now his own company SoftEXIT. James, welcome

00:17 to TechUnhinged. It's great to have you on the podcast. Thank you Robbie. Nice to be here. We see that most companies just use AI

00:24 to build apps. You built the tool that does the building where someone describes what they need in plain

00:30 English and it creates the app for you. What's the real difference between those two? The real difference is everyone is

00:36 using, as you point out correctly, they're they're building apps, but it's generating code, right? And it's also

00:41 generating bugs and security issues and all sorts of things that come with it, right? And it has feature drift where

00:48 you lose features, right? You have to constantly do regression test every time you ask for a build from AI. you're

00:54 getting a whole new application regardless if somebody wants to recognize that or not. So you have to really do full regression tests. That

01:00 seemed laborious to me and basically I created a system where we have a platform and it already exists and it

01:07 you know an individual can go in and customize it and make it to their needs. But if AI were to be the assistant and

01:14 be able to configure all that, it configures it on a stable platform that has been thoroughly tested that they

01:20 don't have to worry about bugs. That the worst thing that could possibly happen is something got misconfigured and then

01:25 they can go in and reconfigure it without having to generate new code without creating more exposure to more

01:32 defects and security issues. So that's the approach we took. It's not generating an application. It's

01:37 generating an application on a platform and that makes it way more predictable and stable. So James, it's been, you know, around 20

01:43 years that you've been in the, you know, software industry from big companies like IBM and McDonald's to your own

01:50 startups. Is AI actually changing how you build software or is it just making the old way faster? What do you think?

01:57 Oh, 100% changed. It's completely different. The previous things that I did on my own, I literally wrote the

02:03 code. I've been writing code since the beginning of my career. I'm really good at it. I mean, I would hope after doing it for so long that I am good at it.

02:10 Now, I don't write code anymore. Honestly, I've become more of the director in this orchestra here. And

02:15 it's very much like when I'm working for a company when I'm not writing code. I'm giving direction to engineers. I'm

02:21 telling them explicitly how I want it designed, explicitly how I want the data to be set up, the communications,

02:27 everything. I actually kind of upset architects because I involve myself so deeply in the design of the system that

02:33 I find that I'm performing that exact same role with AI, but AI becomes the

02:39 coder. That's the real difference. The pace is so fast that every day there is some new feature, new update, and you

02:45 just got to keep up, you know, with all of that. Otherwise, you're probably outdated or you're not going to make it.

02:51 It depends on where you're going, right? If you want to work for a corporation and you want to impress them with your

02:56 AI knowledge, you got to know all the latest, right? You got to know cloud, you got to know uh chatb, you got to know all of that stuff. But it's a

03:02 different perspective when you're building it for yourself. You really don't need to learn all that because technically they're all the same agents

03:09 and all that. They're all a disguise of the LLM. And my theory is now I'm sure

03:14 people will tell me I'm wrong but I really believe I'm right though without any facts is I think that Claude Code is

03:21 what they have is they have the same LLM engine right I mean they didn't just invent this when they left ChatGPT

03:28 overnight so they basically utilize the same thing and then they prompt it and

03:33 you know you have the instructions right you have the skills I think those are all just prompts that you don't have to pay for resubmitting over and over and

03:40 they put a wrapper around it LLM and they're calling And hey, these are agents. These are new. And agents are just iterative prompts. So, it's really

03:46 the same thing. And if you approached it that way, you might not have to pay for all those fancy bells and whistles. You

03:52 can do it yourself on a $20 account in 2026. What do you think that AI, you know, genuinely can build on its own

04:00 from start to finish? And where does it still lag or fall apart?

04:05 So, I think the answer to your first question will answer your second. What What can it build all on its own? Absolutely nothing. I really I think it

04:13 is a great coder. I think it completely lacks long-term planning, long-term

04:19 designing and that I think if you approach something in small projects like you say build me this function, it

04:25 is fantastic at it. If you say build me Salesforce, forget it. I mean it's just

04:31 not going to happen. So it's not going to happen to the level that you won't iterate and iterate and iterate and iterate. So you probably would have been

04:37 better off designing it a piece at a time rather than asking it to build the world then rebuild the world then

04:43 rebuild the world. You know what I mean? That's my impression of it. So you know if we get a bit practical, James, where are you actually using AI

04:50 agents at the moment? What work are they doing for you and what problem are they really solving?

04:56 I don't use agents at all. I really don't. Right now I'm not trying to automate sales leads and all that stuff.

05:01 So maybe I'll come back and look at that. I I don't think I would build those out. I think I would look towards

05:07 the third parties that have already figured all that out and just kind of connect them all together. But for

05:12 building an application, I don't use agents. I literally am down to the bare basics of where I use ChatGPT and I

05:20 have a conversation with it and I build it a function at a time and it saves me all the AI tokens and it gives me better

05:26 quality code, more control over the design. And this is just my method. I'm sure everyone out there is like, "Well,

05:32 he doesn't know what he's doing." Okay, that's fine. But I've built something pretty incredible by doing that. So, can you talk about, you know, what

05:38 have you tried building with it or is it something that you sort of plan to do? Well, actually, I'm probably a month

05:44 away from releasing SoftEXIT Studio, which essentially is a platform, an

05:50 enterprise platform for companies. It's got a dynamic database. It's got dashboards. It's got it's everything's

05:56 very familiar, right? But we have low overhead. So, we're going to come out and offer this at, you know, $14 a month

06:02 per user. So, we're going to really catch it on cost, but, you know, I don't want to go into a sales pitch here, but

06:08 it's way easier to use. It's beautiful looking. It's it's got incredible features that require almost no real

06:15 technical knowledge from people like relational database. All you have to do is say here's the parent and child and

06:20 it's done. Everything else adjusts all the forms and everything. So, all this was built with AI and it was built with

06:26 AI in mind. So at the end that the AI architect that we have built into the

06:32 platform will be able to configure it all because it they speak the same language. There's really no translation.

06:37 It's pretty amazing. There was a study by METR of experienced developers using AI. They thought it

06:43 made them faster, but it actually made them slower and they didn't really notice. When your own team says AI is

06:49 helping, how do you check whether it really is helping them? Well, I think the study is based upon they reviewed

06:56 all the check-ins for GitHub and and things like that. And I believe, you know, we're still in the infancy of

07:01 this. I think they just turned developers loose and said, "Hey, go at it. Here's AI. Make yourself more

07:06 productive with no direction." And they all went in that direction and just created a ton of iterations which

07:12 actually slowed them down. So since they weren't coding, it felt it was faster, but the data showed they had way more

07:18 iterations and in the end they were about 20% less productive than they thought they were more they thought they

07:24 were 20% more productive. I can clearly see how that would happen. I think you need new type of roles like AI

07:31 architects to be able to come in and give them direction on how to use the tool as opposed to just giving them an

07:36 order. make yourself more efficient because I think leadership doesn't fully understand this yet and all they want to

07:42 do is communicate to their their sea level leadership that hey we're doing AI right cuz everyone's got to check that

07:48 box but no one is saying what the directive is of the results of that box being checked there's this old rule in software the

07:56 first 90% of the work takes 90% of the time and the last 10% takes another 90%

08:02 so what do you think AI has made the first 90% almost free that's what they're saying out there. Has it done

08:08 anything for the hard part at the end to make it this? Not at all. I And there's two reasons.

08:14 There's technical and then there's reality, right? AI could generate a perfect application because I know this

08:20 without AI. I have generated many perfect application. You bring it to leadership and you you're going to get

08:26 you know what I don't like this font. I don't like where you put this menu, but it functions perfectly, but they don't

08:31 like it. And getting through that process is an extra 35%. I I guarantee on every project is this constant

08:38 iteration of opinions, right? And you're going to that AI is not going to eliminate that. Leadership is going to

08:44 have their say and they're going to have their opinions. And then you have your customers who are not going to like the way it was done, right? So as you go

08:49 through your focus groups, you're going to go through more iterations of that. You're still you still have the same problem of your BAS are going to give

08:56 you your input. You're going to design it and it's not always going to work out the way it was asked. And you're going

09:01 to still have that stabilization as well. stabilization with bugs, you're still going to have bugs no matter what.

09:07 So, none of that goes away. So, that last 90% is is usually going to take the longest. The saying I'm used to saying

09:13 is, "Well, we're done, but we're not finished." That's that's been my favorite one. [laughter] Yeah, makes sense. James, at Deloitte, you

09:20 did federal work where every requirement was written down before anyone built anything. So, the industry spent years

09:26 trying to get rid of that particular step, but AI can only build what you can clearly describe to it. Are we about to

09:32 need that particular skill back? We hired one. Absolutely. That's that was a really

09:37 insightful point you brought up. I'm telling you, I have been talking to some of my peers and I said, you know, we

09:44 have done ourselves a disservice with agile programming. We've done away with all the documentation and all that. It was great. And then we traded all that

09:50 for micromanaging of developers, right? And everybody thought this was a great idea. I came from also Motorola and that

09:57 was a CMM level 5 organization at Motorola Solutions incredibly wellrun and they had very defined processes for

10:03 software development and I remember the phrase where they would say they would use rational row suite I don't know if

10:08 you're familiar with that but it is IBM should really dust that off and repurpose it for AI now because that

10:14 tool will be amazing so the concept was that they would design the application

10:19 so far that the phrase was the code would write itself and you would have all the requirements they would define

10:25 it right down to the function. So the developers had very little creativity. They would be given they would be signed a function with the inputs and the

10:31 outputs and they would have to write to that function. So all this was designed in rational suite and it was very kind

10:37 of process heavy. I think if we did that well guess what the phrase well the code

10:42 will write itself we have achieved the code will write itself actually. So

10:47 having purely defined requirements fed into an AI engine would be absolutely

10:53 amazing and I really think people should go back and look at the rational row suite and reconsider building those type

11:00 of applications because in my opinion coding is virtually gone. You know that's some of the people who

11:05 are sitting at the top you know that's what they were saying that you know who cares about the code quality anymore as

11:10 long as it's giving you what you want. I throw one thing in there though. I agree because you know when I went from coding

11:17 to management I had a I struggled right cuz not you promote your best coder and your best coder is has high standards

11:24 and it's expecting everyone else to code and at that level and I really struggled with the fact that developers were not

11:29 as good as me at the time right and and this is just levels it's not this is not pride this is just a senior engineer

11:36 looking at entry level so the thing is is the only way you can compensate for that is process and defining your data

11:43 and defining the structure and have them code to it. So if you let them design a bad data structure and a bad

11:49 architectural design, it doesn't matter how good your code is, you can't dig yourself out of that. But if you define

11:55 your data really well and efficient and clean tables, clean relationships between data and clean design of the

12:01 architecture. If the code works and it passes, it's okay. If the performance isn't there, then you have a performance

12:07 problem and you deal with it. But if the code works, it's fine because you can always clean up the code. But you can't

12:13 clean up a design easily because it has a domino effect. So I really think that people need to really start focusing on

12:19 design now and get and don't worry about what the code is being generated, right? And that I'm sure programmers are going,

12:25 "Oh my god, I totally disagree." But the truth is is when you branch into management, you can't control the

12:30 quality of code, especially if you're on a large team. If you have a thousand developers, are you going to code review a thousand developers code? You can't.

12:37 So you have to settle on a few things. You have to settle on process and you have to settle on design. and then the

12:43 rest can fall into place. Does that make sense? Yeah. Yeah, it does. So, you know, James, when you ran a credit card

12:49 processing, the kind that gets audited every year where a mistake is really expensive on papers, it's pretty strict.

12:56 There are clear rules which sound perfect for AI to build in. But is that actually the best place to trust AI or

13:02 or the worst? Well, I mean, AI is just like another coder, right? You know, you got a big team. You again, like I said, you're not

13:08 going to be able to review everyone's code. So what you need to rely on is a good QA department, right? And they're

13:14 the ones that are going to find all, you know, the bad code, the the security

13:19 holes and all that. And then so but now that's becoming more automated and I haven't looked at that, but that sounds

13:25 very promising because if you could build an entire application uh from AI and then you can do a full regression

13:31 test with AI, now you really got something because now you got both ends. You can you can trust it better. And I

13:37 believe eventually AI could test way better than a person because testing is a difficult job because you're doing the

13:43 same thing over and over and over again. And then like at McDonald's um they had you know very strict security

13:49 requirements which we met but we had to hire big QA teams and after a while we would cycle them because they would not

13:56 see the defects in front of them because it's human nature. You keep doing the same thing over and over again you're just going to you're going to not have

14:02 the same attention to detail. And then so we would cycle them where AI won't have that problem. It won't get fatigued. It won't get it won't lose

14:08 sensitivity over what it's seen over and over. But I haven't seen a lot of AI testing suites yet. But I got to ask now that you mentioned,

14:15 did you get free Happy Meals when you were working at McDonald's? I did not. And you know, I'll tell you,

14:21 they didn't even give us a discount on the food because they said the profit margins were so thin. I understand that.

14:27 They did feed us well when they had a meeting, though. So, all right. If a business person can just

14:32 describe an app and have AI build it, what happens to the developer and to the whole custom software business people

14:39 build their careers on? I think software developers need to figure out where they're going. I really do. The more I work with it, the more I

14:45 realize I'm really not going to be hiring software developers and that's a huge impact on the industry. Me as a

14:50 software developer, that breaks my heart. I think the way you asked the question is business people asking for apps to be developed. I think that is a

14:57 formula for absolute failure. There's still other people involved in building software. There's the the architect,

15:04 right? There is the the subject matter experts. There's the technical leaders. I don't think you can do away with any

15:10 of those people. I think they're greatly needed to guide AI into building a

15:15 properly designed system. And I think the only real role at risk here is obviously entry level and obviously

15:22 midlevel and a lot of the worker bees that if I was going to hire a development team I probably would hire

15:28 only one person who is a senior person in each area of discipline. So like

15:34 backend front end mobile I would hire three developers and I'd say here's how you're going to use AI and I would

15:39 instruct it on how to approach it like the way I'm using it. I'd have an architect involved. I'd have maybe a DBA

15:45 cuz the like I said the data structures are far more important than the code now because the data structures dictate your

15:51 entire design. So the DBA has now become a superstar. uh the architect is a superstar and then some of your you know

15:57 lead technical people that could actually pick technologies and I think that's really important now like for

16:03 example I would say what I do is I have it use Java for everything because Java is so fast and Java is no longer an

16:11 issue where I need to find a top-end engineer who really understands it because the code is going to be generated rather than picking something

16:18 that's easier for humans to use now I pick something that's faster and more efficient running because I get an

16:24 incredibly performing application. So, you know, James, there's also a growing argument around that regular SaaS

16:30 is dead or dying. So, how does that hold up to you and how do you think about

16:35 that? So, I've been, you know, as I'm building the SaaS, I have been watching a lot of YouTube and stuff and they're all the

16:41 ones that say SaaS is dead. So, I'm like, I better watch that because I'm building that. I don't think it's completely

16:46 dead. I think you will see a reduction in some companies like you know Salesforce is astronomically expensive

16:53 and it doesn't make any sense. It's just a database and they charge an incredible amount for storage. They charge an

16:59 incredible amount for just storing records and you know anyone technical has got to be scratching their head. How

17:04 do they get away with those prices? But so I think you're going to see companies start to look at that as a way to cut

17:11 costs in building their own items. But there's still you know a high bar of entry for a lot of underserved companies

17:18 right so you say to a company take uh I came from the trucking industry a CDL school the largest one in the country

17:25 and the I started to work with people in this area of uh work and if you go to them and say you know what you could

17:31 build your own system in AI right they're going to be like how do I host that right that seems kind of hard they

17:36 don't know how to do that there's a whole business sector even though it's available to them they're not going to be able to do it Because even though you

17:43 can build the app, you got to host it, you got to deal with the security, you got to do all that, some people are not willing to take that on. So maybe I just

17:49 answered my question why people go to Salesforce is because then it's all on the system and they don't have to manage it and they just pay a lot of money for

17:57 you know a simple database that if they had the knowledge they could have hosted it themselves but they don't. So I think

18:02 AI is the same thing even though for for SaaS right even though they can build it the real question is is will they can

18:09 they and then the other question is should they right even though you can should you I could build my own car

18:15 right but you know why would I do that I could go buy one yeah so also there's a lot of debate around accountability these days you

18:21 know when we talk about AI so when AI builds a business application on its own and it gets something you know wrong you

18:28 know once it's live who's really accountable there Is that any different with developers? If a developer screws

18:34 up, who's really accountable? I It's It's the company. I mean, there's that doesn't change. It's not like I'm doing

18:40 a self-driving car and the policeman pulls me over and says, "You were speeding." Well, I wasn't driving. So,

18:45 it's not that, you know, cut and dry. It's you're still responsible. You know, you designed it. You you built it. You

18:52 hosted it. You said it was secure, right? So, the company is responsible just like they were before.

18:57 That's a really interesting point that you made there. So what if we just forget about the hype for a second?

19:02 Which AI tools do you use personally, you know, for for your everyday work and uh couldn't work without at this

19:09 particular moment and which ones should every CTO or CIO should have their hands on?

19:14 I work exclusively with ChatGPT and I'll tell you there may be other advancements

19:19 and areas, but you know, I keep reading that they're all going to be the same. It's all going to be commoditized and it

19:24 doesn't matter really which one you go because they all seem to copy each other. So rather than spending my time

19:30 or wasting my time looking at every single thing because you're right, it's very difficult to keep up with and

19:35 there's a lot of shiny objects to chase after that I've focused on getting my stuff done. And so I just stuck with

19:42 ChatGPT. I didn't even I don't even use their agents. I I literally just have the conversation. One thing I like about

19:48 ChatGPT is it really knows me and it's a little scary on sometimes and uh it starts to understand my style and it

19:54 becomes like basically I would call ChatGPT my co-founder. I really would I don't need a co-founder. I have this

20:01 tool and it literally does everything that I want it to do and has knowledge that I don't possess. And that's exactly

20:06 what you want in a co-founder is somebody who understands. The one thing I suggest if if you go down that path is

20:12 ask ChatGPT to be brutally honest because it's it likes to please and just tell it to attack your ideas, attack

20:19 your designs and it will make you stronger. Guarantee you and your apps as well. It was very interesting for me James to

20:25 you know you telling that you are still using charge majorly but you know in a world of Claude Code you know it's still

20:32 very rare to find people who are just sticking to charge or probably loyal to charity. Do do you know what scares me

20:37 about Claude Code is my wife will tell you this is actually kind of a funny story when I was originally working with

20:42 ChatGPT I started to trust it and I really wasn't doing a lot of backups and I started coding and then I discovered

20:49 feature drift and it just killed like a feature I spent a month working on and then I lost my temper and I was kicking

20:56 cardboard box I was screaming I go language my wife says what's going on I go this thing screwed me I was so mad at

21:03 it that I couldn't believe it why I don't use like Claude Code. It's gets a feature drift. That feature scared me so

21:09 much that I lost code. The thought of turning my entire codebase over to cloud and it can edit anything it wants scares

21:16 the crap out of me. I mean, I'm like, God knows what it's doing. At least when I'm cut and pasting code in out of a

21:21 chat. I know what it's doing. I know the areas I need to test. I know the areas I need to focus on that I won't take that

21:28 step. Now I would imagine I don't fully understand code because I have limited myself to ChatGPT so I can stay

21:35 focused. But the thought though of in turning over entire source code base to an AI and have it edit anything it wants

21:42 is just very unsettling to me that I I don't think I don't think I have that kind of trust with feature drift cuz

21:48 feature drift is far more painful for me than a bug. Bug I can fix. Feature drift is where a feature disappears that was

21:54 coded and I don't know it disappeared. Then I go to demo it and it's gone. That's bad. How do you deal with AI hallucinations

22:01 when you're working? I don't know. I don't really see hallucinations in code. I think hallucinations are more like if you're a

22:07 lawyer and you ask it to, you know, cite a case and it makes up a case, right? So the code, it either works or it doesn't.

22:13 It's just like working with a developer. He's either good or he's not. You're going to know, you know, James, 5 years out, do

22:19 developers become the people who guide and edit AI or does part of building real software stay human? what are you

22:26 betting on? I you know the industry is creating quite a paradox that no one's going to go into coding. So there no one's going

22:32 to grow their coders, right? So there's no more entry level and the mid-levels will eventually become seniors and these

22:38 and the senior developers will basically become your technical architects at a

22:43 lower level. You'll have your senior architect and then you'll have your lower level architects because they understand code and they understand a

22:49 little bit how to design. So, I don't know where we're going because if once you run out of those people in about 10

22:55 years or so, where are you going to grow people from? Right? I think that's a systemic problem. So, I think right now

23:00 everyone's experiencing a heyday that they can pick and choose who's going to be their lead developers and designers. They can reduce their teams, but they're

23:07 actually kind of messing up the industry in the long run. So, I don't know how that plays out because then you're going

23:13 to start to fall into what we were discussing earlier where the business people are asking for an application and they have no idea about tech and the

23:19 thing creates some god-awful pretty prettyl looking monster underneath, right? And you know, there's no

23:26 foundation to the house and I think you're eventually going to hit that path. So, I don't know. I don't know what happens there. So, long term, I

23:32 don't know. Shortterm, it's crazy. Things are happening. It's wild. But at some point we're going to plateau out on

23:38 the knowledge and either AI has to catch up and really figure that out. We're going to be in a world of hurt. I think

23:43 the people will start to go more the way I did. They'll start to use AI on platforms that they can trust instead of

23:50 having it generate pure code. I really think that's the way to go. It's going to be building blocks. So after all these years, James, what's

23:56 a lesson about building software you had to unlearn the hard way? unlearning is you know I I spoke of it before is when

24:02 I got into leadership is you know I I can't expect people to be great code designers and have this vast

24:09 architectural knowledge right u I'm a full stack developer it gives me incredible advantage I've written native

24:15 iPhone apps Android apps backend front end and my career started in database I've been a coder in dozens of languages

24:22 I have a tremendous amount of experience that I can leverage you know expecting that from developers and even sometimes

24:29 architects is something I had to step back and say, "Okay, not everyone's going to have these skills, right?" So,

24:34 and then the other thing I learned was I'm never going to find someone or well, I mean, they exist, but they're hard to

24:40 find the unicorns who are full stack developers and have all this experience. So, we're still going to have to have a

24:45 team, right? It's I I don't see I keep hearing people talk about the one company millionaire, you know, and I'm

24:51 like, that's possible, but it's a unicorn. It's got to have somebody who who knows all of these things in one

24:56 person and it's really really hard to find cuz I've hired a lot of people in my career and people really specialize

25:02 and it amazes me how specialized they are and how I always felt that pigeon holes them you know I in my opinion so

25:09 you usually know someone yeah I know mobile I know backwards or whatever I said well can I talk to you about the database structures oh I don't know

25:15 about that I just know what I give to a develop a a back-end developer and he tells me the fields I got to get right

25:21 so I think we still need those kind of people. And I guess that's the thing, you know, I had to unlearn and realize

25:27 that we're not going to find all these full stack developers. And that's before even AI is that I had to kind of accept

25:33 that people specialized and still have a fairly large team, right? I still need a UX designer. I still need a BA. I still

25:40 need a DBA. I still need an architect. I still need a security person. I still need a DevOps person, right? And you

25:47 know in theory yeah all these could be done by one person if they have the AI and they have the knowledge but you

25:52 really rarely find that in any individual question it's kind of a roundabout way to get to it but

25:58 so James you know having worked with Deloitte McDonald's huge names on your portfolio when and why did you decide to

26:05 leave it all behind and do something of your own? What problem were you trying to solve? Well I mean I always build something. Uh

26:12 it's it's how I stay current. It's how I'm able to face developers in the room and they take me quite serious cuz I I

26:18 know all the technologies. It's always funny when a developer tries to kind of bluff me into something and I tell them,

26:23 "Yeah, bring up the code." And they're like, "What?" So I go, "Dude, I can code like with the best of them." And you

26:29 know, you're going to show me your code and you're going to prove what you just said to me, right? So I I think that having that coding background is is

26:37 really good. And I always develop something. So sometimes I work for a large company and sometimes I like to

26:43 work for myself and it it all depends and once I finish this I may go back and work for another large company again

26:49 because it's great exposure you know kind of do this on the side maybe or do it full-time. It all depends. I mean if

26:55 it takes off then yes I'm going to do this full-time. If it doesn't then well I go back and work for a large company

27:00 and I learn a lot from everyone around me. So there's benefits to both. Running your own company you got the freedom. I

27:05 can explore any technology I want. I don't have to be held by people's deadlines or perceptions of what an application is. So I have a lot of

27:11 freedom to grow my skills and then working for a large corporation I get to benefit from the knowledge of other people look at larger scale things have

27:18 to work within the constructs of what the company is designing right sometimes you know a little uh forest restrictions

27:25 is you learn a lot from it so you don't have the freedom so you have to follow the rules and that's not a bad thing you

27:30 learn something from it you know another aspect that I'd like to discuss that what change have you been observing or seeing in in the new in the

27:37 new hires you know the um the interns or these software engineers in comparison, you know, to the ones who've been

27:44 recruited way before AI even existed. Are they more smart or they just know

27:49 how to work with AI and lacks their own contextual or knowledge depth?

27:54 You know, I it depends. It depends on where they come from. If if you uh if if

28:00 you're this is really going to upset developers. I know this. I'm sorry, but it's my experience. I've hired literally

28:05 hundreds and hundreds of developers. If if you're a Python programmer, you're attracted to it because probably someone

28:11 said the phrase, "Oh man, it's easy, dude. You don't have to learn all this stuff from Java." Right? And I usually

28:16 find they're not really good developers because it's easy because that's why they went to it cuz they didn't want to

28:22 really work with the challenging stuff. The best developers I've ever seen in my life, and they may know Python, but it's

28:29 the Java developers. These guys are the best, the best of the best, and every time I work with them, they amaze me.

28:35 and they know how to use AI. They know how to make it sing and it it's a world of difference. It's like you could hand

28:42 a Stratavarius to somebody, you know, in high school that, you know, didn't really know, you know, lessons very well

28:47 and and the instrument sounds okay and then you hand it to somebody who is a concert violinist and they play that

28:54 thing and it brings tears to your eyes. And that is the difference of the individual using AI. It's not AI, it's

29:00 still the person. I'm making AI do incredible things because I have knowledge in all these different

29:06 backgrounds, mobile, front end, backend, database, the whole bit. So I can ask it in ways that other people will not ask

29:13 it in their very narrow silos, right? So I think the Java developers are usually more established older developers cuz

29:19 that was the language of its time and it was hot. So I would say they are way more skilled and more more suited for

29:26 this AI world than like the Python developers that are coming because they have never had to work with anything

29:31 really challenging and difficult. So and it's just my opinion, right? And I I'm sure I'm wrong. I'm sure people would

29:37 send me a lot of hate mail on that, but from my experience from the limited developers I've worked with that the

29:44 Java developers have been way better at this game than the Python developers.

29:49 And and I could extend this argument into the Microsoft world because I come from that world too, right? The Visual

29:54 Basic and the C developers, those things compiled at one point to the same exact binary. Two totally different

30:01 programmers. So I'm sure the Microsoft world is facing the same thing that you know, you put AI in the hands of a C

30:09 developer compared to an old Visual Basic developer, you're going to get two different applications at two different

30:14 skill level. Well, James, moving towards the last question. what advice or one thing you

30:19 would want to tell to the CTO's or IT leaders listening to you know get right

30:25 about AI and you know if they should be trusting AI at this point or not. I would give him this advice and this this

30:31 is really interesting because one of the studies as uh that I that I posted on my

30:36 LinkedIn was a study from Stanford University and the first thing the guy said was CTO's all over the world are

30:42 going crazy because their CEOs are busting in their office and saying, "Hey, I saw this guy on LinkedIn. He

30:48 generated his application in a half hour. Why the hell aren't we doing it?" So, you know, here's what I'm going to

30:53 tell the CEO. Okay, so thank you for that. Okay, but here's the thing. Nothing on LinkedIn is vetted or

31:00 peer-reviewed and all of this stuff is like hype just to get, you know, people to look at them. I don't agree with that

31:07 statement that you could generate it in a half hour. Okay? One, I'll tell you, I'll prove it to you why I don't believe

31:12 that. Because if that were true, I wouldn't be working for you. I would be doing it myself and I would go build

31:19 that in an hour and go sell it and then build something else in an hour and go sell it and I could create a fleet of

31:25 applications and become a billionaire overnight. So if I'm not doing that with all my knowledge, then I don't think it

31:30 can be done. So that's the problem is there's this perception that you can

31:35 build stuff in 10 minutes and then you have this massive product. Now that only may be true on my platform, but that's

31:42 not cuz it's generating code. It's configurating a stable platform that already exists. So that could be

31:48 generated in 10 minutes, but you're building that on a base that's been built over a year by multiple

31:54 developers. So, it's a different it's a totally different thing than generating code. So, I guess the the real advice is

31:59 stop believing the hype. Listen to your technical leadership and believe them because they're not making it up. What

32:05 why would they resist doing that? And the CEOs of the world out there can really struggle with that feedback

32:11 because they don't believe it. They're saying, "Well, I got 12 guys I've seen on LinkedIn who do this." They're not. Look into it deep. It's not

32:17 peer-reviewed. They're these guys are not billionaires. You know, they've produced something that looks fantastic.

32:23 I can do that, too. and under the hood it's a mess. No, this was a really interesting and important point that you're making

32:28 towards the end of the conversation. But yeah, you know, that's the ultimate reality and truth around it that yes,

32:34 CEOs get pretty much hyped around this idea. But then they're also coming, you know, to the understanding that most of

32:40 this is just gimmick and it's not really working in reality and hence, you know,

32:45 we are where we are and again it also really depends on whose hands it's in. That's why you

32:51 still need your technical staff and they need to guide this. And those skills do not go away cuz AI doesn't have them.

32:59 Not yet, you know, and I I'm sure I'll get to not yet, but from what I've seen, it does amazing things on a small scale.

33:05 On large, it doesn't do so well, you know, and and to be fair, if somebody came into my office and said, "Build me

33:11 Salesforce." I'd be like, "Oh my god, okay, I built it." And then I show it to them, they go, "Well, that's not what I wanted." Well, your requirements were so

33:18 short that I feel I've met your requirements, but your requirements are not well defined. Right. And this goes

33:23 back to our earlier discussion about how important now requirements are. And the more detail you have, the better your AI

33:29 output will be. So that's my message. Yeah. [clears throat] Yeah. No, 100%. Well, James, thank you so much, you

33:35 know, for this conversation and for your honesty. Absolute pleasure to have you on TechUnhinged with us. Thank you, Rabia. This has been a great

33:41 pleasure. Thank you for having me. [music]