I thought this book would already have been written by someone, and I wanted to see this book so that I could use it as material for an article. In my mind, I was going to write a 250 page book, so for the first time I thought since I'd never published anything, I should probably make it look like I knew what I was doing, and so I created a detailed plan for the book and estimated the page count of the book and it came out to me 900 pages. Co-complete is a single book on co-construction, which includes coding, debugging, detail design, testing, and many more topics. It's also with the most detailed one with impressive 900 pages. The second edition of the book was published in 2004, and 20 years later the book remains the bestseller in software engineering.
I sat down with the author of the book, Steve McConnell in Seattle. One episode we get into, the history of writing co-complete, and how the publisher did not expect the book to sell well. The difference between software construction and coding. Steve's mental model of career development for software engineers, and why he considers the concept of lily pad-hopping, harmful, the impact of AI on software engineering, topics that Steve loved out of co-complete, but now talks about in-depth, and many more. Steve rarely gives interviews, so I hope you enjoy this special one.
This podcast episode is presented by StaffSig, the unified platform for flags, analytics, experiments, and more. Check out the show notes to learn more about them and our other seasoned sponsor. If you enjoy the show, please subscribe to the podcast on any podcast platform and on YouTube. Steve, welcome to the podcast. It's so good to meet you in person.
Thanks for having me. You wrote this book co-complete. I'm going to show it here. I was amazed when I first got my hands on the physical version on how long it is, how extensive it is, and how thorough it is. The other big surprise I had, I assumed that whoever wrote this must have been a really seasoned professional working 10, 20, probably 30, or 40 years, but I learned that you wrote this somewhat early career.
How did you write it? Why did you even come up with the idea of writing such a massive and extensive book? I started out to write a book. I started out to write an article, and I never published anything. I thought I wanted to write, and I thought about writing all different kinds of things.
I was active as a programmer, so I thought right about what you know. I started doing research. I'm a good researcher, which is funny in today's context, but back then, I involved going to a research library and looking through a card catalog and requesting materials through your library loan. It was quite a different process. There was actually some skill involved in that.
Basically, I thought this book would already have been written by someone, and I wanted to see this book so that I could use it as material for an article. I basically started doing research, and after reading about, I think I read about 80 articles and a handful of books, I convinced myself that this book didn't exist. This was just baffling to me because there were books on requirements, design, testing, project management, but there wasn't a book on the main thing the programmers do, which is software construction. I just kind of shifted years, and I didn't set out to write a 900-page book, which was the first edition. I thought it's funny because in my mind, I was going to write a 250-page book because all the other books I'd read in that space had been 250-350 pages.
I did background research. I wrote a couple of sample chapters. I got ready to submit my proposal to the publisher. At this point, I only had two chapters written. For the first time, I thought, since I'd never published anything, I should probably make it look like I knew what I was doing.
I created a detailed plan for the book, and for the first time, I estimated the page count of the book, and it came out to me 900 pages. I thought that can't possibly be right. I'm not writing a 250-page book. I estimated it a different way, and I came up with something like 875 pages, so I thought, okay, I think I'm writing a 900-page book. I think the thing about that is that if I had thought a year earlier that it was going to be 900 pages, it never would have occurred to me to even try.
By that time, I was too far into it. I was mentally and emotionally committed, and so I decided to just go ahead and do it. Just help us imagine what stage of your career were in terms of work experience, college, those kind of things. I was about five years out of college at that point. I'd been programming professionally for six years, I guess, because I took some time off in the middle of college.
I actually think that was the perfect time to write the book because I think it's really important in writing a book to have a really clear sense of the target audience. For me, my target audience was me five years earlier. It was recent enough that I could still remember where I was five years earlier and why I didn't know five years earlier. I think for somebody who was 20 or 30 years into their career, it'd be really hard to know. Plus, you'd just be out of date in terms of what you needed to know.
After the book came out, what was the impact of it? Did you remember initially and then a couple years later? Yeah. The book came out and it started to get really good reviews. At that time, the reviews, of course, were in magazines which people got in the mail on paper.
They trickled in, way pre-Amazon. The reviews were great. I attended a book signing here in Bellevue and one of the people from Microsoft Press came in and said, oh, this is our sleeper book. I learned that they never expected it to actually sell very well, that they decided to publish it because they thought it would be good for the catalog to have a more scholarly in-depth book, but they assumed it wouldn't really sell very well. The sales have been a nice surprise for everybody.
You must have learned about a really interesting inspiration to the book had on the blog called Coding Horror. I don't remember exactly when it was. I don't remember. 2005, 2005, or something like that by Jeff Offwood. He started the blog or he renamed it, I'm not sure.
I think he conceived it from the beginning as Coding Horror. He reached out to me and said, hey, I'm writing this blog or I want to call it Coding Horror. Can I have permission to use the Coding Horror icon from your book? And I was like, yeah, sure. And I think he sent me a t-shirt or something with the Coding Horror image on it.
And of course, his blog did really, really well and led to a lot of good things for him and I think had a positive impact on the world too. I think at some point, it might have been the most popular developer blog. He published that about something like 100,000 requests per day. I was reading it. So when I was early career, I read Coding Horror almost like every day or like every other day he published an article.
And then of course, that led to Stack Overflow or that helped him meet people, Joel and others and that led to Stack Overflow. So it's just kind of incredible how indirectly, obviously, but your book inspired someone else to start a blog which inspired or helped create this very special site, Stack Overflow, which Stack Overflow all these now has helped train some of the agents. It's just all trickling all down. Yeah. Well, I think, I don't know that I believe in karma literally, but I definitely believe in the general idea that if you put out good things into the world that there are positive ripple effects and you never know what exactly is going to happen, but probably more good things than bad will happen.
And so it sounds like that's part of what happened with that. And so the book is on the concept called software construction and you know, you kind of started the book, I'll quote a little bit from it. You say at one time software development and coding were thought to be one and the same, but as district activities in the software development life cycle have been identified, some of the best binds in the field has spent time analyzing the meeting methods of project management, requirements, design and testing. They rush to study these identified, but it left code construction as the ignorant cousin of software development. Let's ask, what is code construction?
I mean, the book is about code construction, but I think it helps us recap and how is different to let's say software development or how similar to it. That was pretty deliberate in the book in trying to capture the space of software construction and a space that's distinct from writing code or not distinct from, but is not the same as writing code. I think that the idea of just jumping and encoding is something that maybe you do when you're first learning how to program. But as you get a little bit more advanced, I think there's a level of intermediate thought that includes thinking about how you're going to test what you're writing, thinking about how you're going to design in the details, what you're writing, actually doing the coding and the full set, the universe of topics that come to bear on the actual code itself, the readability and all aspects of readability.
I think that much of the literature before I wrote Code Complete, there were a couple things out there that were decent things on coding style, but that was kind of it. It was coding style. It wasn't really about, we had books on software design, but the software design really stopped short of the level of design thinking that you would get into at what would have been one time the routine level and then later was really the class level. It just seemed to me like there was just a lot of thought that wasn't necessarily guided or necessarily taking place in terms of thinking about the activities that are adjacent to coding, but not quite big enough to qualify for a book or a name of their own.
Then just to be clear, these laws are clearly, obviously we write the code, but it's everything that encapsulates until I get something that is ready to ship, that is ready to, literally we have the planning or requirements or business requirements. Then it starts from me thinking about how I'm going to structure my code as I build it, I create, I debug, I check that it's running, I think about longer-term intended. These are all part of code construction. I'm just double checking that. I'm understanding because I feel that code construction in the book, you definitely covered, but since then I don't think we talk about code construction that much.
I would probably make it a little narrower than what you just described. I think that some of what you just described, I would probably characterize it as a higher-level design. But on any given project, depending on how the project evolves or how it emerges, the design level of thing could happen at the beginning or could happen incrementally as you go. It could happen by somebody who is good at that kind of thinking or could happen by the people who are actually doing the implementation. I think one of my learnings over the years, and I didn't really understand this when I wrote the first edition of Code Complete, is really that different people's brains work differently.
Some people are really capable of thinking about design abstractly and then implementing the design. Some people are really good at writing the code and learning the lessons in the small and then building up from that. Some people are good at both, but I don't think a lot of people are actually good at both. I think much more commonly you find people who are good at one or the other. I think there's often a disconnect between the people who have ability in one area and not the other.
I've seen cases where people think that it's just not possible to work the other way because they're not able to do it and they just can't imagine that somebody else could be able to work that way and yet we've got all kinds of instance proofs of people working in both ways and having it work out fine. And then the two ways I understand is one of the two different ways of working. You could think of it generally as top-down versus bottom-up. So you could think of it at where inductive versus deductive. I believe that most programmers, meaning the majority, not like 99%, but 51% or maybe 70%.
I believe most programmers are pre-detailed oriented. And so I made a conscious design decision as I was writing code complete that I would write the book inductively. I wouldn't start out making a bunch of general statements and then demonstrate the specifics. I would start out talking about a lot of specifics and then build up to the general statements at the end. I just felt like more programmers than not were more oriented in that direction.
But what I didn't appreciate at the time I was writing that was how big the disconnect was. And I think a lot of people just aren't capable of operating the other direction, whichever the other direction. Steve was just talking about how there are different ways to get to the good design of a software system. Some developers do a top-down, others do a bottom-up. But how do you know if a well-designed system works as expected in the real world?
This is where analytics and experimentation are so important. And it's why companies like Meta, Uber, and others build complex experimentation systems themselves. These are far more advanced than what engineers at most companies could have access to. Unsurrecently, that is when Statsik became available. Statsik is our presenting partner for the season and they built a complete set of data tools that allow engineering teams to measure the impact of their work, focus on outcomes, and create bottoms-up products culture.
This includes advanced A-B testing, product analytics, feature flags, plus tools like session replays, a user store, and more. With Statsik, every time you release a new feature, you can see exactly how it moves a product or information that you care about. And then you track progress over time using the same data set. This tool can still be valuable to so many teams at OpenAI, who is a huge user of Statsik, decides to acquire the company recently, talk about validation. Statsik has an incredibly generous reach here, a pro tier for teams starting at $150 per month, and scale of enterprise pricing.
To learn more, go to statsik.com slash pragmatic. Yeah, this is really interesting because, you know, for example, these days at a lot of scale-ups tech companies, it's becoming increasingly expected of engineers to start with a design document, which is a software design document. And when I work, for example, at Uber, we operate like this, but a lot of times it just feels like a nag. I know what I want to write, which makes me realize, I think, your points on some people are different. They're always the same developers or engineers who wrote the code, did the thing, and they're like, okay, I'll write a design document just to check the box with the code already written.
And at the time, I tried to explain to these people, especially eventually I became a manager to some of these people. You need to start, you know, highlight your high level thing. And I didn't appreciate it. I was like, what? Because I guess maybe I am fine operating like this.
I'm probably in this first group. I didn't appreciate this until now. Yeah. There's a great paper, really old paper by Dave Parnas called the rational design process, how and why to fake it. And the gist of the paper is from, I think, the mid-70s.
I don't remember exactly. It was a really old paper. But the gist of the paper is that design is a sloppy process. And nobody gets it right. It's not linear.
You don't start out and march your way in a straight line toward getting a good design. And so number one, the paper sort of gave permission to experiment, try stuff, make mistakes, learn from the mistakes, iterate. But then, then, the paper also says, how and why to fake it part was, at the end of that, you describe what you did. And the description is not necessarily going to show you all of the mistakes and dead ends and so on. So it's going to look like it was a more rational process than it was.
But so I think both halves of that paper are worth keeping in mind. And it sounds like your experience would kind of bear that out. It's not even really possible to get it right the first time in most cases. Yeah. And I do sense that in the industry these days, we've kind of forgotten a little bit about the design or the iteration or even things like, I have to remember who said this.
I think it was John Alsterhout who in his book on software design, a philosophy of software design, he suggested, well, design it twice because the second time you'll get a better design than the first time. These are just things that some people might still do. I would say three times, but yeah. And I'm serious about that. So when I was at Microsoft, there was a programmer inside Microsoft who was famous at the time who was well known for rewriting everything three times.
And the basic gist of it was, if you had two weeks to do an assignment, he would spend the first six days coding and making all kinds of mistakes and not doing it very well. And then he just throw it out and spend the next three days rewriting it. By that time he would be applying all the lessons he's learned. Then he spent the last day or day in half writing a third version and it would be perfect. And if you think about your own experience, if you've ever, you know, this should never happen in theory, but it still happens where you've done a bunch of work and somehow you lose it.
It gets deleted, you'd have a backup, whatever, right? It should never happen, but sometimes it does. And you think, oh, this is awful. I just spent the whole day working on this. It's going to take me forever.
And then you rewrite it in 45 minutes. And as you go, you remember all the lessons that you learned and all the stuff you wish you'd done the first time. It actually turns out probably better than if you'd never lost it in the first place and had to keep what you wrote initially. Well, these days, I think it happens a bit less than to get being everywhere committed. But I've had this earlier in my career.
And as you said, first of all, you blow up. You're like so mad. Like it could, I think it was a file loss or deleted or something like that. And as you said, you're thinking, oh, it's going to be so much work. I don't even want to do it.
But when I now remember the same lesson, I spent some time on the school board here in Bellevue. And one of the surprising learnings I got from the school board was that one of the best ways to build reading proficiency is to have kids read things they already have read. This repetitive reading builds confidence and builds skill. And so I think you kind of apply the same idea of rewriting the same code you've already written that maybe there's a lost opportunity there because we're always trying to do something new and we don't often get the chance to go back and just read you the thing we just did. Yeah.
And I think this is one of the things where I wonder if we're talking a lot about software design in the industry because we just ship once. We don't even go back to rewrite the whole thing, especially with web-based software, you know, mobile is becoming that as a you can ship it. You can change it any time. You know, there's startups, lean startups from 2010, gave this concept of MVP, minimum buyable product, you know, it doesn't need to be good. It just needs to be good enough.
And I guess the most mature companies, which hit so-called product market for the product that works over time, they will do a rewrite or maybe a secondary write through a series of migrations and all these choices. So they will get actually a lot better, but that's kind of it. So I think there's a lesson we can probably all just consider. That's not a bad thing to rewrite. So your career path, we know that you wrote this book early career.
What was your professional development story up to the book and then after the book? What happened after? Up to the book, I had worked for a small insurance consulting company. I worked for a large, briefly for a large insurance company. I had worked at Boeing on a defense project.
I had worked in a startup mode for a company that was kind of doing a just a startup shrink wrap product. And I was about to start working on the book full time and I got an offer to go to Microsoft, just initially for a summer. And I thought, oh, at the time, Microsoft was the hot tech company. And so I thought, yeah, if I get a chance to see what's going on inside Microsoft, I should do that. It'll make my book better.
So I ended up being at Microsoft for a year. That was essentially my background. I also done, I always had some kind of startup project of my own going throughout that period. And I put a lot of time into those as well. I had started out being kind of interested in programming, but not super interested in it, notwithstanding the fact that I seemed to spend a lot of my essentially hobby time on it anyway.
And so I started out just kind of thinking, I wasn't really sure if I wanted to do this as a job. And then I thought, well, but I had a little bit of entrepreneurial streak. So I kept doing startup stuff in that mode. And at some point in there, I kind of switched over from thinking, I know if I really want to do this as a job to thinking, if I'm going to do this as a job, I probably ought to figure out what I'm doing. And I think at the time, for the first couple of years, I had a little bit of imposter syndrome, because I didn't have a degree in computer science.
I didn't know what the time that I needed. Oh, really? Of course, it was. What was your degree on philosophy? Yeah, I had minors in math and computer science, but at the time, less than half the people working in the field had degrees in computer science.
That didn't change for a long time. So I didn't realize that that was common. I thought that was unusual. So I started trying to develop or acquire some skills so that I could basically be more professional. So I started going to user group meetings.
I had one friend who was as interested in the programming stuff as I was. We go to these things together and talk about them. And then I got the chance to be, I worked at Boeing and I was like, okay, I'm not seeing a whole lot here that really looks like qualitatively different professional programming to me. And then I got inside Microsoft and I said, okay, I'm finally going to see what real professional programming looks like. The best of the industry, like Microsoft back then was a bit like what OpenAI is now or Google in like 2004.
It was just the 800 pound gorilla and there was nothing else even close at the time. And I don't know what percentage of all software share of the all market software share they have. So that was around the peak, right? Around the windows and DOS and that error, right? Windows was not really popular yet.
That's what I worked on. But it was sort of the last phase of the DOS era. Which one of those was dominated? Beginning phase of Windows, yeah. And they weren't super dominant in applications the way they are, the way they are now.
So I got inside Microsoft and what I discovered was number one, people were really really smart. So that was a factor. And number two, you had like one person working on the printer device driver for one specific printer. So the difference between that and my experience was my experience after that point had been you're working for a company and you've got responsibility for figuring out how to get things to print, figuring out how to load and save files, figuring out how to display stuff on the screen. And it's like, oh, I've got like a whole person not just on printing but on printing the one device.
So I mean, that was a lot of it. It's just the projects were broken down into much, much smaller pieces. And so people could put a lot more time into it. And then yeah, they were really, really bright. But qualitatively, in terms of programming practices, you know, there wasn't really anything different than what I'd already seen.
So you got it's Microsoft. You write the book. And this point, how did you think about your career? I think you had a talk earlier, which is not recorded, but about the mental model on how you viewed your career at the time and then figured out what next. At the time I wrote CodeComplete, I was still trying to decide if I really wanted to be programmer at all.
And if you can believe that. And so I spent a year full time writing CodeComplete, but I spent two years doing background research before I spent the year full time. So you did research on like, you got a full time job and on the side you were doing. And then you like took off time. I took a year, a year off.
Wow. The dot not looks scary. I mean, suddenly you were there with, you know, like no job just writing this book. Yeah, I really wanted to write the book. That was the main thing I wanted to do by then.
And I knew it was a big project because then I figured out it was 900 pages. And with the 250 page that I'd been carrying around my mind, that could have been maybe a part time, but 900 pages, I knew I needed to take some dedicated time to do it. And I wanted to do it. I'd save money and I could, you know, I could afford to take some time off. I had basically, I was living a very low overhead life at the time.
I didn't own a house. I wasn't married. I didn't have a dog, you know, so I could live really cheaply if I needed to, which I did. And so I spent about a year writing full time. And at the end of that, number one, it was a terrific experience.
At the time I felt like I'd gotten three years of experience in one year by writing the book. I would say I learned so much more about what I was doing just through the process of the writing is only not the main part of it. The thinking about what I wanted to write was the main part of it. But then I went back into hands-on programming role and I felt like, oh, this is great. You know, I like the writing, but I'm so glad I'm doing hands-on programming because I was kind of sick of the writing at that point.
But by the time I published Go Jumpleet, I was convinced I was never going to write another book. And because the end phase of that was just painful. And so then I worked for a couple years and the memory of the pain faded. And I was kind of getting the itch to write something again. So then I wrote my second book, Rapid Development.
And I kept thinking, I was going back and forth. I kept thinking, do I like writing better or do I like programming better? I kind of liked whatever I was doing at the time better. It took me a while. Eventually I realized I liked the back and forth.
I liked the variety, but not the variety on a day to day, week to week, month to month basis. I liked having big blocks of time to concentrate and dive in, go really deep. So before I wrote Rapid Development, at that point I was starting to think, all right, yeah, I think this programming thing might be a good direction for me. And maybe I am a little committed to the career. And so I started to think about where do I think my career should go and what do I want to do.
And so I have no idea how I came up with the idea, but I created what I thought of as the career pyramid. And I just kind of organized the software universe into the base of the pyramid was all programmers, like 100% of the programmers and my focus was on US. And then the second level up was programmers who had maybe actually gotten a degree in programming, which I had not. And then the next level up was maybe advanced degree in programming or maybe written some magazine articles or were in a leader manager position. And so anyway, the pyramid kind of went up to where there were just a very small number of people at the top who had essentially a lifetime of significant contributions to the field.
And so I thought, okay, just in terms of setting a direction, I'll just aim for the top of the pyramid. And it wasn't so much because I thought I wanted to get to the top of the pyramid. It was just more to have a point on the horizon that I could move toward. And so I use that partly to guide what I did on the second book because I had a variety of books. I was interested in writing and the book that would have made the most sense after code complete would have been design complete.
I could have written it at the time, but I felt like I would get pigeonholed as just a pure tech guy if I did that. The other book I really wanted to write was performance optimization in Windows, which was a fun topic, but again, it was a little bit too bits and bytes. And so I decided to hop over and write something that was a little bit more management oriented because I wanted to kind of triangulate on the top of the pyramid. And by the way, this pyramid you shared a book, so well also shared, so I thought it was such a good or interesting mental model that I haven't seen many people think. And I really appreciated that you kind of zoomed out and looked like, all right, where am I?
Where would I like to go? How do I go there? As part of this, you mentioned again in this talk, you mentioned the concept of lily-pat hopping. What is that and how did you apply it to your career? Well, I think that the pyramid was, in essence, the antidote to lily-pat hopping.
What I had noticed, I noticed it a lot more later, but even at the time I'd realized that programming didn't really feel like a career. It felt like a job. And the reason it felt like a job rather than a career was I didn't get the sense that it was going anywhere. I saw people who'd work on one project and they'd work on the next project and then they'd work on the next project. That's the lily-pat hopping part.
But it didn't add up to anything. It was just different. It was kind of like difference for the sake of difference. And you know, they're probably learning things. They're learning a new business area.
They're learning new technology. And you know, maybe that's interesting and stimulating and there's definitely some personal growth that goes on there. But I always felt like if I'm going to put the effort into learning something and growing, I'd like it to add up. I'd like it to accumulate. I don't want it to just be a bunch of different things that are all the same in a way.
I really wanted to be essentially more than the sum of the parts. Part of the reason for me coming up with the pyramid in the first place was really just trying to think about, okay, how do I get past this idea of I've worked on this project, I've worked on that project. I know a lot more stuff now than I used to, but I'm no more marketable. I really know more valuable. I don't have the ability to really contribute on any individual project or organization more than I used to have.
So how do I essentially build my skills in such a way that I become increasingly valuable as time goes by? You're going to put in the same effort either way. And so it's almost back to the design question where it's like, if I'm going to put in this much effort, I can either have it add up to this or I can have it add up to that or that is a lot more than this if I put some thought into it. So I really like this thing, especially because as developers, I think we forget our software engineers that we have a lot of freedom on the job. For example, when I was a manager to engineers, if an engineer came to me and said on my team and said like, Hey, I'm really interested in this area.
First of all, if they said that I would be thinking, how can I help them with that? But even more so, but sometimes that would be tricky to do or not as important as to say like, Hey, I'm interested in this, this, this other project. I'd love to work on it because I think it would help my skill set. So if they kind of did the work to just look around, understand the limitation of the team, whatnot, like my answer was like amazing. And I also thought, wow, this is a person who's really taking initiative.
Let me help them because as your manager, I think, you know, I'm not sure if people still have the notion of like the other, like the boss who's a who's something negative, like as a manager, you really want everyone to thrive. You want people to grow professionally because guess what? They'll have better output, more motivation. It makes your team better and as a manager, that's your goal. So I think the more people realize that you can think about where you want to go, even at the workplace, you can almost always make some of it happen, if not all of it.
I think that's right. There's saying that most people live in a box that they build for themselves. And then Larry Constantine is one of the founders of Structured Design, if that term even means anything. Had a saying that I like which he says, nobody needs to give you permission to do your job well. And the core layer to that is your managers already think that you're going to do whatever needs to be done to do your job well.
They assume that you believe that you have permission to do it. Meanwhile, on the ground, a lot of people live in that box they build for themselves and they think, oh, I got to do this the same way that they did it last time. There's no reason they think that is just like, guess it's human nature for people to just kind of constrain themselves that way. So I think you're absolutely right that in the vast majority of software jobs that I've ever been aware of, people can design a lot of their own job or their own day-to-day work in such a way that they can learn a lot and grow a lot in a more strategic way. And I think this is just becoming, it's starting to become more of a baseline expectation.
So one thing I noticed is when we were hiring people at Uber, Uber, we had a culture that was very start-up-be experimental. We had a saying be an owner, not a renter, do what needs to be done. And I have people who joined from like, consultancies, places where it was common for the client or someone to sign tickets to them, for example, Jira tickets, and they would work on it. And they got used to it. They weren't really allowed to go and ask questions.
It wasn't kind of like, don't question it. And so when some of those people came over, they found it really hard to adjust because they were like, oh, you know, what tickets should I sell? And I was like, well, like, don't tell me. You tell me. But it's interesting how that's the first time I saw that there was a divide where people at some point in their career, they got told, you know, like, this is the box.
They inside the box. And then they found it really hard to, oh, I can actually, there's nothing here. I can actually go outside. My first summer job that I got paid for programming is an internship. And I basically finished all the work they had for me halfway through the summer.
So they had to make up stuff for me to do. And my boss gave me the assignment. He just handed me a box of floppy diskettes. You have to put up a picture so people know what those are. And he said, I want you to do something useful with this box of diskettes.
And my response was, well, what do you want me to do? And he said, I don't know what I want you to do. Part of your job is to figure out what I should want you to do. So you take this away and, you know, do something useful with it. And at the time, I was incredibly frustrated.
I felt like the boss wasn't delegating anything to me. He wasn't giving me clear direction. I really felt like he sort of abdicated his responsibilities as a boss. As the years went by, and certainly by the time I started my own company, that was basically the model for 99% of my day, which is nobody, nobody, when you're the owner of your company, tells you what the next most useful thing to do is. You got to figure that out.
In fact, it's really mainly your job to be the one who's looking at the whole landscape and saying, what do I do next? What's the next most useful thing to do? So yeah, I completely agree with that and identify with it. And it's funny how, as a junior person, you can feel like the ambiguity in that assignment is a real liability, but really that's incredibly valuable training for what you deal with as you get into higher positions of responsibility. And when I say responsibility, I'm not just talking about for people, I'm talking about product direction, health of the organization, all kinds of stuff.
Just to see how baseline is, I do know if we were exceptional or not, but just a few weeks ago, I talked with someone at OpenAI about, in fact, I talked with the head of engineering at OpenAI, and they're now one of the hottest companies right now, a bit like how Microsoft was in the 90s or Google later, or maybe Uber even at some point. And I asked the head of engineering, how are you structuring your team's engineers? Is that right? You told me something interesting, how they don't really have roles anymore? Because they actually have some of the designers write code, obviously, with the help of AI, some of the engineers do the product requirements or even do design.
And it's all becoming really fluid. And he said that the model they follow is everyone should pick up both what they think is the most important thing to do. And so suddenly, what used to be a pretty well-defined role of software engineer, product manager, designer, they're all kind of murky. I mean, people still have their expertise. And this is one of the most innovative companies right now.
I'm seeing servers do more and more of this. They're now calling it actually product engineer, which means you are a software engineer, but your job is not just to write code, your job is to understand what we do. And then build solutions for the most amazing problems. Now, it is spreading across that more and more startups are only hiring for these folks. And if you're used to being told what to do or you're looking for it, they'll be like, sorry, we're not going for these types of folks.
Yeah. We just talked about how open the AI works differently than many companies, even in the roles they have. One other thing there are different is, in the tools engineers that the company used to get things done. They use linear, where the seasoned sponsor of this podcast and the big reason for their choice was they need to harmonize all the different ways their product team was working. Picture this.
You're an engineer working on a critical project. You need to coordinate with another engineering team on a shared dependency. But here's the problem. Your team tracks work in JIRA. They use GitHub issues.
And the product team writes specs in Google Docs. Every handoff requires you to translate not just to work, but the entire way of thinking about the work. This slows you down massively. What open AI engineer described as an archipelago, where each team is on their own island using their own language, which sounds funny. But from my experience at Uber, it's super accurate.
So their special linear was driven by the need to establish a shared vocabulary between teams. Every project now follows the same lifecycle, uses the same labels, moves through the same states. When an engineer from one team needs to understand what another team is working on, they don't have to learn a new system or decode different workflows. Their usage of linear in this way is how they can move so fast, despite being quite a large company. And to close, linear is just so simple to use.
Their team needed no formal training. Try it for yourself by heading to linear.app.pragmatic. Well, what comes to my mind when you say that is, I think companies go through a maturing process. When I say maturing, in this case, I'm not even saying more mature is better than less mature. Sometimes less mature is better.
But you have certain personalities that work well in startup mode. And you have other personalities that work better as you get into slow growth or sustaining mode. And so what you described, I think, is great. When you've got most of the staff that's self-motivated, self-driven, and they're part of something exciting and it's growing, and they're probably young and they're probably hired directly out of school or a couple years out of school, there's a lot you can do in that mode. Because the company gets bigger, as it gets installed claim-based, as other companies start to do things that rely heavily on the reliability, not just the innovation, but they need different things.
They no longer need what's latest and greatest and coolest. They need the thing to work the same way it did yesterday because they've made assumptions that depend on it. Then you get, essentially, you're calling for a different kind of staff to work on that. So as time goes by, some of the staff that were the biggest contributors early on get bored and frustrated with all the restrictions and they leave. And then all that ambiguity or flexibility in the roles and stuff, that's no longer a positive.
So my prediction with this scenario you described is that 10 years from now, that organization will need what my company called a roles and responsibilities workshop because it's just so common that you start out in a way that that works. But you get to a point where, as things scale up and as the characteristics of the staff change, it just doesn't work anymore. And so you just need to be a little bit more explicit about it. And I think these are all just natural pendulum swings that happen as time goes by and it's not that one is better or worse than the other. It's really fit for purpose, which in startup mode, having a lot of bureaucracy in startup mode is a great recipe for never getting started.
But adding the right kind of bureaucracy at some point becomes necessary to sustain an organization of a larger size. I think I think I very much right, especially because it is popular now. Obviously I also talk with a lot of basically startups. Sometimes they're scale up, but they're still very young growing, especially in AI. It's a new topic.
And if you look at companies, there's this joke that goes into tech industry, eventually every company becomes IBM. Yeah. And now like when you look at Google in some ways, well, they're behaving awfully a lot like IBM used to, but if you look at Google is now 30 or well, it's going to turn 30 years old. It's a massive business. It has 100, 200,000 people, et cetera.
Like as you said, the things don't work. And one model I've heard, which I don't know who to attribute it to, but there's the pioneer settler and city builder, which also is one variation of what you sent. And I had age on top of that. When I was at Microsoft in the early 90s, the campus age was aging 0.8 years per calendar year. And so when I was at Microsoft, I was 27.
Average age on campus was 27 point something or 26 point something. I was about half a year off the average age on campus. And that's average. That's not media. And that's average.
And so if we've got one guy who's 60, you need a lot of guys who are like 22 to get the average down to 27. And so the average, you know, misrepresented how old the campus was really. It was probably younger than younger than younger than the media must have been younger. Yeah. So when I was there, there were about 15,000 people total and the Dev staff was about 2000.
So but aging point eight demographic years per calendar year, fast forward 15 years. Now the campus is 40, not 27. And now people have got houses and they've got kids and they've got spouses. And you know, it's just all this other stuff that's more on the kind of technical, technology, code base, customer base, there's all the business stuff. But I think the demographics of the staff matter.
And the reality is that people in the company age as the company ages and that has some pretty big, pretty big impacts. Yeah. And some of the companies that have been the hottest and fastest growing startup is all the time and thinking of Amazon, of Google, of meta. You know, like their CEOs and their staff is now, you know, it used to be 20 when they started or even I think Facebook 19, something like that. Now they're in a 40, 45, 50 as you say, it's all growing with it.
Yeah. It's good to keep in mind. Now also in this kind of career advice talk that you did, you outlying three areas that you've noticed that software developers should become proficient in technology knowledge, business domain knowledge and software best practices knowledge in your experience. Like how important are these and how can you get good at them? And if you have to, you know, take a stab at which ones are becoming more important as a developer is getting more experience with in your experience, which one was, you know, like one more important leader for you or the people you work with.
They all matter to some degree at all times. If you're a junior developer, you're not very good at all of them or any of them, but you're probably the strongest in the technology area. If you're a little bit less junior developer, you know, maybe you've worked for the same company for five years, then you could be pretty strong in the software best practices or pretty strong in the business in addition to being pretty strong on the technology side. And five years is a long time in the software world. At some point, the vast majority of software professionals I've ever met, they end up concentrating in one area or another.
And so I think the most career limiting, unless you're in just a super tech organization, is to just go all in on the technology side. There are some places where that's going to work, but that's the exception. Going all in on the business side is a really good background for moving into more of the company level architect role. You understand how the different pieces fit together. For both the technology and the business, you're able to make technical decisions on a business basis and not very many people are good at that.
And so I think that can be super valuable, useful career path. The best practices path, which is my path, is actually probably a little bit more limited in the vast majority of companies. It's less limited if you're a consultant or trainer. But inside a company, you can still be valuable, but you're probably going to move into more of a coach role or possibly move into more of a manager role. Anyway, I think either of those second focus areas can be viable and valuable paths to focus on.
And I think one thing I mentioned in that talk was that it's very useful to think through, which one do you want to be? And I don't know that a lot of people actually think about it that much. They get an opportunity to be a manager. And so they're like, okay, that seems like a step up. And they don't necessarily realize, oh, yeah, if I take the manager path, part of what goes with that is I need to get better at the people side of it.
And then I also need to get better at the business side of it. If I go high enough on the manager path, at some point, the business needs to come into conflict with the technology needs and my role is going to require me to favor the needs of the business. And a lot of technical people can't make that transition because they're too wedded to the technology and they just can't conceive of the idea that the business would take precedence over the best, you know, that's in quotes, the best technical decision. Yeah, that's interesting with the manager path. I think we've had about 10 years where looking back that we had zero interest rates and companies were growing really fast investing.
So, I think the business side of the technology was huge. Thanks to the mobile revolution, smart for revolution, thanks to cloud, thanks to startup returns and IPOs, which is being big. And going to the manager path used to be seen as a very low risk one as one where you could go and you could manage people, maybe make a bit more money, open your options because now you could go to found the startup, say I have management experience, you could go back to being an engineer later as well. But what happened now, well, some of those people who thought, oh, I'll just go into management for a little bit and see how it is and I'll go back. They ended up going higher and higher again with this high growth area, earnings potential was higher.
But now the reality is hitting where teams are not growing as fast. In fact, there are some companies scaling back. So, managers are finding themselves formerly technical people in this really awkward position where they're not that technical anymore. Their past few years of experience is scaling fast growing teams, which is no longer needed. So I've had some friends who saw this coming a little bit earlier, head of engineering or director of engineering and some of them went back to individual contributors or to tech lead roles where technically they took a step down.
But some of them are actually happier for it and it just gives a lot more stability. There's still a lot of engineers who need to be hired, but just fewer managers and especially fewer managers without a warm referral. Yeah, I don't have enough knowledge of sales staff or general business staff to know how it looks on that side of the fence. But from the reading I've done, I feel like technical staff have a couple of interesting attributes as they move into management. I don't read about in general management literature.
One is that I think a pretty common pattern for technical staff is going into management for a year or two, deciding they don't like it, going back into a technical contributor role, enjoying it, but going back to thinking about, huh, here could have done the management thing this way, it could have done it that way, and then they make a second attempt at manager and then they actually really like it and they do well. I've seen that pattern many times. And so just the idea of kind of bouncing up against the ceiling, coming back, regrouping, and then taking a second run at it a few years later. And some of those people end up being really, really effective managers. Another attribute that I find to be really uncommon is that most technical managers at any level don't have any aspiration to general management positions.
If you're at VP of technology role, VP of software role, it's unlikely that you aspire to be the CEO or the COO. What's likely if you aspire to anything is you aspire to having the VP technology role at a cooler company. And that's it. And so it's funny because I think for general business people, this doesn't really compute. They're like, hey, if I'm a VP, I want to be the CEO.
I want to move into the C level. But for whatever reason, I don't think technical people think that way. I don't read about it. I was in technology. Yeah.
And there are some really interesting stories as well. The founder and for a long time CEO, Hashi Corp, Michael Hashimoto, he was the CEO of the company for a while. I think he needed to, I think he took over from his co-founder and then he went back to being an individual contributor at the same company, which is uncommon. But it comes to show and now he's actually, he went back to writing, you know, he founded Hashi Corp by writing software. I think Terraform or some of the others.
And now he's back to writing software. It just shows that he just loves building. Yeah. You know, there's a great leadership book called What Got You Here Won't Get You There. And the gist of it is really aimed at organizational leaders.
And the gist of the book is here are all the things you did to kind of get to where you are. And they were all great for that role. You're not in that role anymore. And so it'd be successful in your new role. You have to operate differently.
And some people don't want to operate differently. So the ones that actually have some self-awareness do what you describe. They go back into a different role. I think the specific scenario you described of the founder of the company going back into a contribute role, organizationally, we've seen that as super messy just because it's hard to disagree with the owner and all that stuff. And it can be incredibly, incredibly randomizing for the organization.
But if that same person instead goes off and starts the next new thing, that can be great. You know, they can think of everything they learn the first time. I do think that a lot of tech founders don't really think or even talk or consume literature or learnings from other companies. You know, there's a lot of like first-timers. Why not?
There's some drama and conflicts and learning sometimes works out sometimes. It doesn't. But I wonder if this is engineering specific or any company has this tendency of like, look, I know what I'm good at, which is engineering and you assume that you're going to be good at other stuff as well, which sometimes you are. Sometimes you are. Sometimes you're not.
Yeah. Well, I think, you know, a topic that I don't think I've ever written about, but that I think is worth just throwing on the table is that focus the application of personal energy makes up for a lot of other deficiencies. And in startup mode, when you're talking about not having well-defined roles and so on, if everybody's actually working in a focus way and they care about getting work done and they're applying a lot of energy, you know, mistakes aren't a big deal. You just go in and you fix it because you've got the energy to do it. If you're working in a super-processor anyway and people are kind of thinking half about their home life and they're checked out because now they're 45 instead of 27, not that that's a universal pattern or anything.
But, you know, if you don't have the same level of personal energy to correct mistakes, the mistakes linger, they affect more people, affect more downstream products and it just becomes a much bigger deal. And, you know, when I was at Microsoft, the core Windows team was under 15 people. And of course, this was, you know, much earlier version of Windows, but at the time there was a project going on at IBM, well, Microsoft and IBM were jointly developing operating system called OS2 and IBM had something like 10 times the staff, maybe more than that. And Microsoft was really frustrated at how slow IBM was on their side. And the reality is you can get a lot of stuff done with the right 10 people who are all super-focused.
And even if you just are thinking about it in terms of number communication paths, how many people do I have to interact with on a day-to-day hour basis? I've got nine or 10 people. How many does it take to make a decision that sticks? How many does it take to fully consider an issue? You know, I hear if I have a decision or a 10-person team, I need four people to weigh in.
I've now heard everything with four people. If I've got 100 people on the team, I've got, you know, 30 that need to weigh in. That's an entirely different proposition. And so, yeah, you can get an awful lot done with the right, with a high level of energy. And I don't think in the technical space, I haven't read much at all on people talking about the role that energy plays.
It's all about process and training your different people and technology and so on. And if I just think back of the past, like, just 20, 30 years, the ones that I've seen repeatedly, regardless of technology, regardless of, you know, right now we have AI earlier, we have like co-generators in terms of tools or in telecents that just auto-complete your thing. You know, we have all these tools that make teams productive. But whatever tools we have, if I think back from the 90s all the way to like today, we always see these things where there's a five, 10-person team comes out of nowhere, builds this thing. And so, you know, I think in the gaming, there's the thing that built its software, you know, they built Doom with, I think, like, three core people, nine people, and it was a huge hit than in social media with Instagram, which was 10 people, Facebook acquired them because they were growing so fast, they would have been bigger on Facebook WhatsApp.
Again, it was a small team, but I think there were 50 people by the end when they got bought. Right now we have a social media platform called Blue Sky, which again, they're with 12 engineers. They're going really fast, open AI when they're small and so on. And it feels that it doesn't really matter what the tech stage is like what connects them as, well, energy focus. Yeah, and I've not read anything about this.
You know, there's always the interview people and like, how did you do that? And they tell you some kind of story. And when you ask people about like, so right now I have been talking with like some people, for example, opening I just right now, they're hard as well. I asked them like, what is it special? How do you ship so fast like open AI still out shipping some smaller startups as a larger organization?
And the answer I get sounds cliche, but they're telling you to say, well, it's the people and how everyone's so focused and motivated. They don't say energy, but yeah, that as well. Yeah. And so during the era, I was at Microsoft, I think the culture Microsoft when I was there was if you were awake, you were working. And the thing that it looks from the outside like it's a sweatshop, but from the inside, it's just they were doing an incredible job at that time of hiring people who really didn't want to do anything other than work on the next generation of software.
And it was a great time to be there because you felt like you were at least a year ahead of the rest of the world because number one, you were developing it at Microsoft. Number two, everybody else who's doing anything interesting wanted to come to Microsoft to show it to you. And so you were constantly in one mode at work where you were seeing the future. And then when you went home or talked to somebody from a different company, they were a year or two behind where you were. So it's incredibly exciting, stimulating place to be.
And of course, you just naturally wanted to spend all your time working, but you can't replicate that through good management where you're just trying to somehow motivate people to want to work 100% of their waking hours. You really have to just put the pieces in place so that people want to do it. And if you're a super high-marky company like Microsoft was at that time, then it happens kind of automatically. Really, the manager's job is not to mess it up. But if you're a regular, just business or a sustaining mode company, I don't think it's possible to replicate in those companies.
Every once in a while, you'll see some individual project where somebody does a half-a-decent job of replicating it for a short time. But yeah, so all the examples that you gave resonate with me because they're all kind of in the same mold of super exciting time to be there, part of something, you're creating the state of the art, advancing the state of the universe. Unfortunately, it doesn't last forever. And so it's cool to be a part of it while it lasts. Yeah.
And I think those of us, for example, I've had opportunities to be part of a team where at the time we weren't changing the universe, but we were doing something really exciting and felt we're up against the world. It felt a bit impossible as well. Those are relationships where I still keep in touch with some of the people. It's like friendships are made through these crazy times. And it's weird because I wanted to, it is crunch time, but a little bit of self-inflicted, sometimes it's external pressure, but I would never advocate for that.
But I'm also very glad that I did it. I do want some times in my life like that, not all the time, because that's burnout, but also if you have none of it, it's just nine to five then. I'm 100%. Yeah. And it was funny because we saw the agile movement go through the same kind of maturing of its view of sustainable pace that my company had gone through.
We originally, as a company really before agile existed, had started out thinking 40-hour work week, but the first guy that I hired who ended up being the CTO for 20 years, eventually he and I realized neither of us had ever worked in a 40-hour work week. We both naturally worked in burst mode and we both did our best work in burst mode. And so it's not about having a steady state. It's about having a sustainable pattern. And what you described resonates with me 100% in that you don't want it all the time, but yeah, if you had to go through your whole career and never experience that, I would really feel like I had missed out on something if I couldn't have been in that mode at least some.
I came up with this analogy, which is imperfect, but I used to do sports at a competitive level and this involves swimming and running and especially running because it is something where you can be injured. When you're training, you have when you're actually sprinting or you're doing the competition, but even in training, you're doing that part. You're also doing the endurance part where you're not getting all that. And then you're also resting. And a coach will instruct an athlete to go like, okay, go all out, give me 70%, give me 50% rest for this much.
And if I think about our professional career, if I had a coach who was looking to maximize my impact on the long term, athletes only 10 years for us, 40 years, they would actually tell me the same. And yet, when we're in a job, we feel bad about coasting a little bit after a hard project, but we probably should be doing this. Maybe don't advertise it because there's a bad stigma these days when you say like, oh, I'm doing it. But after you do something really tough, etc. It's okay.
Yeah, as long as that's not, you're not getting stuck in that, right? Yeah, I didn't really spend that long at Microsoft, but for some reason, it keeps coming back to that topic. And I think the being part of that is exciting for a while. As you say, when it becomes a marathon and not, or becomes multiple marathons back to back, it doesn't really work as well. And Microsoft in that era had a lot of stuff going on.
It wasn't really that exciting. And so there's this background culture that people have participated in where they're working themselves to death. And after a couple of years, they're like, I don't really want to do this anymore. And so there were cases where as the years went by, there's sort of developing expectation that that was going to be the work mode. We're now in a completely different space when there's an expectation that it's going to occur rather than just the individuals doing it because that's what they want to do.
And Microsoft, because they were so successful, got into a mode where they basically had to go to people and keep products and say, you know, we need you for one more release because we've identified that. We need to carry forward at least two of the key people from the last release to be successful in the next release. And they were basically saying, look, we'll give you enough stock to make you rich, but we need you one more release on this. And so now that's just a completely different vibe on the project than the one before. So you mentioned that one topic you didn't write about, maybe it could have been a good idea, a high-sized rule of energy.
I wanted to ask you, since the second edition has been published, that must have been 20-something years as well, what are two other topics that if you wrote this book today, you might have added into it. Yeah, so I think that really gets into the limitation of where I am professionally now. And I really have not been active in software at the code complete level for several years now. I mean, the irony of it is I've spent more time actually doing hands-on programming just for my personal use the last five years than I did for probably 15 years before that. But it's really just one person project, me personally doing it.
I think we've talked quite a bit in this conversation about design three times. I think one thing that has worked out really well for code complete, second edition, is really that it is the second edition. And then I published it 11 years after the first edition. And so I think that was really kind of an ideal amount of time between the two editions, because I learned a lot in the intervening 10 years. The industry, I think, had matured quite a lot.
And when I went back and started working in the second edition, I had a great vantage point for looking at, okay, what has changed since the first edition and really thinking through, well, what insights does that give me about the kinds of things that are shorter lived? And the whole point of the book was to try to capture the longer lived best practices that would span generations of technology and span programming languages and so on. So the 10 years between those two, there was some stuff in the first edition that's like, yeah, okay, this really wasn't lasting. This needs to go. There was other stuff.
It's like, wow, you know, it's been 10 years. This really hasn't changed much at all. And I think when the second edition came out in 2004, Agile was really, it was beyond infancy. It was kind of at the toddler's stage at that point, maybe a little beyond that. But we were still mostly focused on XP.
So I really hadn't emerged as the dominant practice at that point. And so I think one of the challenges for me in the second edition was trying to figure out, okay, well, how much do I really want to talk about Agile and how much do I really want to talk about XP in particular? I decided really I only used XP in one example in the book. So I'm pretty happy with how that worked out. But in the first edition, that consideration really been about object oriented because when I was writing the first edition, object oriented was well established in academic circles, but not well established in practice.
And so I made a decision to really not do very much with it. And in the intervening 10 years, it became so well established that it wasn't really a separate topic. I originally thought I'd have some separate treatment of it, but really it just ended up weaving its way in really throughout the book. And it didn't make sense to try to treat it as a separate topic. So I feel like the first edition had aged more in 10 years than the second edition has aged in 20 years.
Yeah. And I actually like I just collected a few topics. We don't need to go through. But the things that I thought did not age or actually age well, so it's very applicable. The parts of managing complexity, you go through how you can use things like abstraction, clear conventions to reduce cognitive load.
Our brains have not gotten bigger in the last 20 years. No, cohesion and coupling, how cohesion within modules, but also lose coupling between the modules, health for more maintainable software also has not changed. I want to ask you, in the book, it goes into the value of iteration and continuous improvement. Was this, because I feel this is like, dah, like this is how we work. But at the time when you wrote it, was this a bit less accepted still or a bit more novel of any other, because it's interesting to see things that were like today, like obviously, but these are not how ideas start, right?
Yeah, no, it's a really interesting question. And it does take me back to kind of the state of the world in 2004. In 2004, the reason we were talking about incrementalism and iteration had nothing to do with product delivery. It had to do, it really had to do with efficient technical practices and a virtuous learning cycle. But kind of around 2004, we started to go from internet brochureware to actual meaningful, more rich functionality websites.
So the idea of CICD wasn't really a concept in 2004, but we had done work for Amazon in about that same time frame. And at the time AWS didn't exist, at least not by name. But what we'd observed in the work for Amazon was I thought they were okay at the technical programming level, but I thought they were probably the best in the world at operations, as we called it at the time. And their concept of just, if it doesn't work, we'll roll it back. They just weren't a lot of organizations back then that could do that.
And so, yeah, in the last 20 years with mobile, well, internet at first and then mobile, and just functionality getting rolled out continuously, these two, what's good for the learning cycle and for containing errors at the development level has merged very nicely with how you all deliver functionality, yeah, the product. Yeah. And so, you know, today, like when I was writing my book, the software engineering's guidebook that was published two years ago, one topic I was really unsure of, should I include or not, a little bit to your agile thing was JANAI, because chat GPT had come out about a year before, like before when I finished my manuscript, like six months before, and you could already see that it was big.
It was great for learning. It was good at generating code. So I figured, should I add this or not? Because in the end, I added it because it felt to me already all the developers I knew were already using it, even if just to explain things. And since then, one of the biggest changes, how JANAI is becoming, I'm not going to talk about the generics, but for coding, it turns out it is very good at generating code.
And for example, in the book, one technique that you encourage and what I've used before is the C-O-L-Cote programming process where you come up with C-O-Lote code. And for example, when you do that, it's great at giving it the more detailed C-O-Lote code the better and you can tell it generates Java, Kotlin, Rust, whatever. It does a great job of doing so. So I just want to get your thoughts and your first principles now that I know you've not been as embedded in software in the past few years, but knowing that there's this tool, which is very good at generating code, it's a weird intelligence. It's not intelligence for say.
It generates an ex-token, but it does it really well. And what we're seeing is it spits out code really, really quickly, oftentimes good, sometimes not. I think some people, like I compared when I was talking with John Oster, I was the tactical tornado of doing so. And there is some scare with some developers of how it's actually coding better than I do. If you were an early career professional in this industry, seeing that there's this tool that is actually good at spitting out code, not perfectly, but pretty good code, how would you think of making the most out of it for your career, to understand how to tame this and how this might even change software development, if it won't, maybe it won't.
Yeah. Well, I think it's pretty clear it will. I think it's totally unclear how it will. In my mind, we're still in the pretty early stages of how we would use any kind of AI for help with software development or really for help with anything. I think at this point in time, the lessons I feel like I've learned are the more guardrails you put in place, the better result you're going to get.
There was a funny LinkedIn post yesterday from somebody who said the learnings about using AI and software are, you've got to be really clear about your requirements, you've got to really define your test case as well, you've got to define all the exception cases really well. And the gist of it was, huh, this just sounds like good software development fundamentals. And only we are now, you know, AI is the thing that's finally after 75 years forcing us to do this. And maybe it hasn't been 75 years that it's been 60 plus. Yeah.
So, I mean, I think that that's funny, but it's also true. Yeah. One of the things that I think is true of software developers is developers are so tech oriented. If you give them a tool to do something, that works a lot better than giving them some rules or guidelines. And so if AI is a tool and now to get the tool to work, you've got to go through these software fundamentals.
Yeah. Maybe that is the thing that finally makes that happen. My other feeling about it, and I would love it if you correct me, if you feel like I'm off base on this, my other feeling is that I think writing the gold path or the happy path through the code has always been the easy part of coding and understanding which exception cases did I forget about or was I unaware of or did the requirements never capture in the first place. Fred Brooks published a paper called No Silver Bullets and he talks about the difference between accidental complexity and essential complexity. And a lot of the essential complexity comes from the real world because the real world is messy.
And to the degree that AI is 100% lined up with the messiness of the real world, then the programmer job of being the party that fully vets all of the exception cases and the corner cases and the off by ones and so on. That's the other thing that's concerning is I still think AI can't count. And so off by one errors seem like they're just probably. So I have to agree because I think we should not forget like obviously the interesting thing with AI, I talk with the really experienced engineer Simon Willis and he created the Django framework. He is a very productive engineer for about 20 or 25 years, just like codes a lot and writes really good code.
And he's been using Chad GPC and all the other tools since they've come out so coming up three years now. And he told me because I asked like, hey, Simon, is it worth understanding of theory? Because I took time to understand how this thing works. It's just matrix everywhere, which is why GPUs work so well with them because turns out 3D operation is also this matrix of multiplications. So it's a huge matrix that are hard for us to understand and they predict the next token.
And he said actually it's it might be a little bit harmful in his experience because you need to get a feel for how this works because now it does work again. He says that it just takes a lot of time to figure out where it works and where it doesn't now in his case, I think for a lot of us more experienced engineers, it can be super helpful because as it works, like I will call BS, I'll be like, this is bad. And it often gets things really bad. So as long as I just use it as like this, people think about it, I think Ken Beck said he thinks of it as his genie where you ask you, it grants your wish, but sometimes it really funny ways. Like in really nice, for example, you tell like, oh, make this work and it does or make it work and the test should not fail.
And sometimes it will just go and delete the test when it cannot. Early in my career on the when I was a Boeing, we had a CDC cyber supercomputer. The way the cyber supercomputer was explained to me is it's a very powerful, very literal giant that will do exactly what you tell it to do. And but it's so literal that it has no ability to do anything other than exactly its interpretation of what you tell it to do. I kind of feel like AI is in that same space is incredibly powerful.
And you know, your comment about sometimes is glaringly off. I don't worry about that case. I worry about it being subtly off. Yes. And it is off subtly a lot.
So there are some already some small stories that are going viral, which are it could have happened to human, but a good example to start up, you know, they use one of the many agents, which does stuff, it created a PR, it looked good. And you know, it's after a while when it generates good stuff, you're like, looks at me, you commit it. So it just made a small change that added up observably logging to one of their things that was unnecessary. And this agent cost $500 a month in a month that generated an extra $800 for nothing. And the guy was like, Oh, okay.
And I think we're going to see a lot of these things. I think we'll have to ask ourselves again, like what is software engineering? Because now what I'm I'm hearing for teams that are using more of those AI is reviewing code is super important. Yeah. But of course, it is hard to like, I remember when I again, I worked at Uber is like the more you got to know an engineer, sometimes you realize they're really reliable.
So eventually I can code review and we needed to two thumbs up or two people to accept. And when that person came and wrote their stuff and it was a long pull request, you tended to say LGBTM looks good to me without reading it. And of course, the shorter it is, we also started to have this rule. I don't have long pull requests because like for a 500 line change, most people were like looks good to me because I just don't have to talk about it for five lines. You get like 50 comments saying, Hey, what about this?
What about that? Et cetera. So it feels like we, I don't know when, but I think we're going to come back to some of the earlier learnings on software engineer because this thing will not change whether it's a machine doing it or whether it's a human doing it or whether a machine that is acting up able to take on some things that a human can do won't go anywhere. Yeah. So sometimes programmers have difficulty just getting started.
You mentioned the pseudocode programming process. I think that's something where AI can be pretty helpful. You're starting to say, Hey, look, what are the steps I should, what are the different topics I should consider when I'm designing this? Okay. What sequence would make sense?
You can basically dialogue and say, you know, give me a breakdown of this space and show me what the class design would look like. Okay. I don't like that. Give me a different breakdown. Show me what that would look like the ability to iterate super quickly and insert your own brain into those iterations, I think is incredibly valuable.
And for that matter, it's incredibly stimulating to be the human in that, in that loop. And then of course you can also say what test cases should I consider, you know, you can kind of pit it against itself. And, and I think at least my own experience with that is, you know, I haven't tried anything super complicated, but I think the general concept seems to work. Yeah. These are definitely also one other thing that seems to be very useful is explaining things.
May that be a language framework technology. Obviously you always need to double check that some of them are already giving references. Yeah. So like, like it can speed you up a lot and help you. You know, one thing I am seeing is things happening across the industry as engineers are going to be more full stock in the sense of you might remember the time where they was like Java engineer and dot that engineer and they didn't mix.
And then later at my time when I started to work, we now had back and engineers work across languages, you could do multiple and I was kind of expected, but still back in the front and mobile were separate. And now with the help of these things, you know, like a back and injury, you can generally mobile pull requests and so on. Yeah. And you know, one of your questions are one of the threads has been kind of what's changed over the years. I think what you just described is a huge change where first edition of Code Complete, a lot of readers would have learned one programming language when they read the book by the second edition.
Maybe, you know, maybe it was more common or more universal for people to have learned two or three, but the idea that it's just assumed that you're going to learn a few because you need to be using a few on a day to day basis and your job is a full stack engineer. That's just a complete change. And I think the mental development that goes along with that of understanding what's common, what's different across languages. I mean, that's just really valuable. Yeah.
And I also wonder if it will be really valuable now because I feel you know, the breath keeps increasing, but don't forget, you know, back in 30 or 40 years ago, the people who learned one language, they were just as smart as the people now. I guess the difference is so they probably started equal knowledge. They needed to tell way more details or a lot more this and that. So I wonder if it'll be more really valuable. We now have a lot more breadth for sure, but you cannot have as much depth.
So the people who are able to balance this and even though you have access to this, you know, gene that can explain everything, actually taking the time to understand and have that mental model, you'll probably go further. Yeah. I think, you know, in thinking about the Fred Brooks, no silver bullets paper, which is pretty interesting to think about in terms of AI because the whole premise, the whole concept of a silver bullet is any single innovation technology development method that by itself is capable of producing an order of magnitude improvement over a 10 year period. You know, AI might very well be the thing that that is the exception to the rule in that paper. But you know, the basic argument is that the real world messiness and other sources of essential complexity aren't going away.
Unless you find a tool that can deal effectively and utterly reliably with all those details, the corner cases, the off by one cases, all that stuff, you know, programming, we don't get to be approximately right. We have to be exactly right. And so, you know, that's really, I think that to me, the gist of it, if AI can get to the point where it's exactly right, then, you know, maybe we really do start to replace programming jobs with AI, but the whole history of programming as languages have gotten more powerful and more expressive, tools have gotten more powerful and more expressive. I think that the defining characteristic of a programmer is being the person whose job it is to figure out what is exactly right.
And if AI just basically changes the playing field, so now you're trying to identify what's exactly right with AI generated output, to me, it's almost the same job it always was. And don't forget that AI generates code and that code at some point needs to be understood, especially because they can get into loops, you know, again, they can get confused because again, just how it works, almost we understand that. So there's valuable both in that. But also, I feel like there's this thinking as well. I'm not sure I subscribe for this, but how, you know, we used to have assembly as the programming language machine code and there was C or C plus plus, then we have compiled languages, we had we have some visual languages or visual basic.
And this can be seen as like yet another layer where English translates it except it's not a one on one mapping like all the other ones. So it's a lot more thing. We'll see where it goes. It's it's changing things. But I think the history of programming has been layers upon layers of you added abstraction, more people can get started.
I mean, a global apparently was invented to make it a lot easier for developers to get started. Everything was invented to eliminate programming. All the programming was invented to eliminate programming. So I would take what you say and I would look at it from a slightly different angle, which is I think the history of programming is a history of increasing levels of aggregation. So you started out with assembly or machine language and then you had aggregation into assembly and that was one level of aggregation.
Then you had aggregation into macro assembly. So I was a second level of aggregation. Then you had this amazing invention called the subroutine, which was an aggregation of callable functionality. And then you had an aggregation in the classes where you had a combination of data and functionality. And then I don't know, it kind of, you know, at least that's going to wear my knowledge of it, my direct involvement of it kind of.
And then now I think you have a combination of like these reusable functionality or like things that, you know, like some of the, I can definitely spit out things that I'd seen a lot. Yeah. And actually we even had like some some templates. So you know, we'll see how it goes. But I like I said, I think it makes sense.
You know, and maybe it is truly that above, you know, there are lots of attempts to have the next level of aggregation way back when Ada had a notion of packages and C sharp and notion namespaces and you know, there have been other attempts and maybe there's an environment out there that I don't know about, which is entirely possible that has done a great job of aggregating packages with their own interface and their own behavioral rules and so on. But you know, if, if AI ends up being the way to achieve the next level of aggregation, that kind of makes sense. Well, maybe, but you know, don't forget one thing that I think we just didn't need of us mentioned, a huge difference with AI and everything else that came before it is non-deterministic. Yeah.
And in programming, we have been used to, and I think we've gotten really good at making the most of the deterministic and non-deterministic is always has its, its dangers, its flaws, its, its harder to work with. Yeah. And often, especially when you're non-deterministic system and you want a deterministic outcome, you need to do a bunch of work around that. So maybe that'll keep us busy for a while. And sometimes it's not possible, by the way, as closing, you've had a long and successful career, you've written a book that actually gets started a lot of people's careers.
I had one of my colleagues from one of my, my companies tell me that in their first, first few years as a developer actually served as a military and they read the book and it really helped them. What suggestions would you have for an engineer to have a durable career principles that, that worked for you, probably independent of technology? When I was writing the first edition of the book, one of the things I tried to keep firmly in mind was that the guiding principle for writing the book and the principle that answered every question was, does this make the book more valuable for the reader? And so there were some things I kind of liked, but maybe were idiosyncratic to me. I ultimately decided that doesn't make it more valuable for the reader.
So I left them out. Designing them into the book, content of the book, everything was about what makes this more valuable for the reader. And as I iterated on that over the course of writing the book, I think that iteration, with that in mind, just added up to something. And I think the same thing with my career advice would be exactly the same. What makes you more valuable to your organization or to the world at large?
And it's not like anybody, I would think someone's a bad person if they don't think that way. But if everything you're doing is just to just thratch some person, it's idiosyncratic, I'm interested in it, but it doesn't add up to something. Just be aware that that's a lost opportunity. And if on the other hand, every time you think about, well, what project should I do next? What do I want to do next in the organization?
Sometimes you don't get a choice. Sometimes you do get a choice between option A and option B. Does option B open up more doors, possibly than option A? Well, maybe you should choose option B for that reason. Does one of them make you more valuable?
Choose it for that reason. And I don't think this is just about personal development or personal success. I think it's about making the world better. If you're making yourself more valuable, you're increasing your ability to contribute to the world. And I think that's a virtuous thing.
And you close the book with how you emphasize how craftsmanship is important for professional growth and you write how curiosity, continuous learning, are essential traits of great software engineers. And to me, it feels just as relevant as ever. I just think it's part of the human experience. And I think that programmers, probably more than average, are curious people who like to learn, who are attracted to novelty, who want to learn what's new and what's cool, and who want to explore what's possible with what's new and cool. A lot of our conversation today about AI has been all about exploring what's possible with what's new and cool.
I think this is just the way programmers are and I wouldn't change that. Steve, it's been both a really good conversation and a privilege. Thanks so much. Thanks so much for having me. It was such a blast to introduce Steve.
After the interview, we grabbed dinner at one of his favorite local restaurants. They told me more about why he decided to make a shift and link to software industry after a successful 30-plus year career to try something different. Steve's energy levels, positivity, and thoughtfulness run deep. And I hope this came across through the podcast as well. For more observations on how the software engineering industry has changed in the last few decades, check out Really Deep Dies at the Private Engineering, which are linked in the show notes below.
If you enjoyed this podcast, please subscribe on our favorite podcast player and on YouTube. This helps more people discover the podcast and a special thank you for negotiating. Thanks and see you in the next one.