How Will We Train Developers If AI Does the Routine Work: A Conversation with Scott Hanselman Transcript, AI Summary & Key Points

Postgress shouldn't force you to choose between transactions and analytics. Tiger data creators of time scale DB adds hybrid row and columnar storage 95% compression and continuous aggregates. So both run in one Postgress database. Try it free at tigerdata.com/go/trial. Welcome to the Architects podcast where we discuss what it means to be an architect and how architects actually do their job.

We are very lucky today to have as our guest Scott Hanselman who is a programmer, teacher, and speaker. He works out of his home office in Portland, Oregon for Microsoft GitHub as a vice president and member of the technical staff. He works on open-source and agentic coding tools like GitHub Copilot CLI along with the Windows development platform. Before Microsoft, he was the chief architect at Corin Corporation.

Now a part of check free for six plus years. Previously he worked as a principal consultant at Step Technology speaking writing consulting and very much not getting rich during web 1.0 although he worked on a bunch of websites you probably have heard of. He also worked at Intel, Nike and consulted all over the world. He was an adjunct professor at OIT teaching C and web services and created the first Palm Pilot diabetes management system in 1998 and sold it to a healthcare company some years later.

It's great to have you here on the podcast and I'd like to start out by asking you, were you trained as an architect? How did you become an architect? It's not something you woke up and decided today I'm going to be an architect. Uh, you know, I mean, architect is one of those titles that one could probably just give themselves, like CTO, like you'll meet a CTO at somewhere like a big Fortune 5000 company or a Fortune 500 company and they're an architect or they're CTO and then you'll meet someone who's at Panera with

their buddy and they're a founder and they're a CTO as well. So, these are titles are kind of made up, but I am trained in software engineering. I do point out interestingly enough that I don't have a computer science degree. I have a software engineering degree which is the practice of shipping which is different than compiler theory, computer theory and the things that one learns in computer science many many many years ago.

I helped write some of the tests for the Microsoft certified software architect stuff with Clemens Vasters you know low these 20 25 years ago. But like what is an architect? Is it someone that knows UML or knows how to do Vizio? Are they the one in the room with good taste? Are they the software somalier? I don't I don't know. >> Maybe with AI we'll get the PowerPoint compiler and that will be >> we definitely have the PowerPoint compiler.

It compiles crap though. But I have said recently on podcasts that AIS generate features extremely well. They do not generate architecture very well. they tend to be somewhat myopic in their view and they definitely miss the big picture. And I've I'm currently working through a a problem with an application that I'm working on where the AI has over the last several weeks created a god object just a superstruct that I would not have created but it's it's kind of a cancer that's building inside the application and I need

to get it out of there before it collapses under its own weight. So the AI did a very poor job of designing an architecture. >> That's interesting segue into something I know we wanted to talk about is that you and Mike Rousinovich wrote an opinion piece in the communications of the ACM last April on redefining the software engineering profession for AI.

And in it you discuss the difficulty of training new software engineers when AI is doing all the work that we traditionally trained entry-level software engineers. Can you explain what you meant by this and the culture of being a perceptor at scale? So the idea is in any internship, apprenticeship, you know, master student, teacher student scenario, often the student is doing the, you know, wax on, wax off work.

They're doing the toil. They're doing the tedium. And it is through the wax the car that you learn karate. And if that job is now done by AI, then at what altitude, as I mix my metaphors, does an early in career person enter? Do they just skip the boring bits? Is it important to do the push-ups and pull-ups and stuff? And I think that it is important.

I think that we are skipping the easy bits and it is extremely dopamine positive and very exciting and it's certainly a thrill. I'm watching my early in career and non-technical co-workers get very very excited creating software and vibing stuff. But without trying to gatekeep the work, I don't believe that you can vibe into production, at least not yet.

I feel like early in career people need guidance. They need a mentor. But the term mentor or or you know apprentice, journeyman, like these are kind of outdated. So Mark and I looked towards successful industries that are regulated and managed and create successful seniors. And we found that in nursing. My wife is a nurse and she is a nurse preceptor.

And a preceptor is an interesting word that is almost exclusively currently used in medicine to refer to the teaching nurse. And that job is a explicit and specific job. Not every nurse is a nurse preceptor. And by having that level of formality such that the nurse preceptor, the nurse trainer is metricked and their job performance is measured on whether or not they can make nurses.

They can mint new senior nurses. In software though, like if I joined your team and you're the CTO or the architect and you're the boss and you make me the intern or the apprentice, upon entrance, I'm less than and then I have to dig myself out of the hole that you've put me in. And then if I don't don't hack it, you'll go, "Well, handsome couldn't hack it.

We'll grind them up and spit them out and we'll get another one." So then interns become a dime a dozen. And if they can't crawl their way out of the bucket, then you'll go and find more. But with preceptors, their job is to make more nurses. So maybe if we couldn't make more nurses, it it's on the the trainer, their annual review be like, maybe Michael shouldn't be a trainer.

Maybe he needs some more training to be a trainer. So there's a formality to it. So instead of a tech lead with a bunch of plebs working for them, you have a tech lead working alongside a preceptor trainer. And that trainer's job is not to ship code, but to ship engineers. So like after this podcast, I'm going to have my scrum and we're going to cut a release of the product that I'm working on.

And there'll be a very mixed group of young people and and uh later in career people. And we're going to do some pair programming and we're going to do some fellowship and some working on stuff together to make sure that we are not just shipping the thing. We're going to cut a release at noon, but that we are shipping senior engineers. That's a big long answer, a lot of rambling there.

But I think you understand that the idea is focus on making people better not just shipping more crap. So two questions come to mind right away when you say that. One is what then becomes the role of software engineering education in the universities >> if this is the preceptor and also does remote work contribute to the problem and the lack of mentorship or does it actually help as someone who has worked remotely now for many years my entire career at Microsoft has been remote I would say that that is hard for anyone

and it's even harder for an early in career. So I would argue that remote work is not helping grow people particular like I've got young men in my house 18 and 20 and they're going to go out into the world. There is value in them shaking hands, looking people in the eye, talking to human beings. And I know that this might sound like two old men on a podcast shaking their fist at the cloud, but human beings are meant to hang out together.

We are meant to be social. I got a lot of value sitting at whiteboards with senior engineers having them explain to me how things work. If you are an online person who does their socializing online, does their video gaming online, does their everything online, does their dating online, and now your work is online, then that shows a fundamental collapse of the third place, right?

Work, home, and the third place. And if there is nowhere, if work, home, and the third place are all the same place, I don't know how you can grow into a fully formed and healthy individual. So yeah, remote work is a challenge. We can push through it, but it requires a lot of metacognition and a lot of focus on listening to your body, listening to yourself, making sure that you are a whole person.

I do drive up to Seattle every 4 to 6 weeks to make sure that my badge still works and that the young people are still still present. Although I do have folks that are remote in other places. It's the perfect storm of co remote work, social media addiction and AI that threatens to break the young people and I can only focus on on one of those. >> So I want to get back to universities in a minute but this is this is interesting.

What you're saying requires acknowledgement on both sides of the equation because it's the young people as you say who live in this virtual world that have to learn the importance of the inperson and you also need the companies to understand that in order to be profitable in the long term they have to be willing to have these preceptors to help people along in the office if possible.

Yes. So, you have to acknowledge that hitting the ground running is not a thing and that an early in career person fresh out of school is not going to start slinging code within, you know, a couple of days. We do have a metric in our group and we call it engineering thrive where we call it time to first PR. So, here's your laptop. Welcome to the company.

It's Monday. And it's not a productivity metric. It's not, oh, Michael joined the company on Monday and he got a PR landed on Wednesday or Thursday. That's actually a morale metric. Think about how exciting it will be for them to land their first PR and how enabled they will be. So when we talk about engineering thrive, we talk about how do they feel in their personal growth.

Even if it's a oneline change, it shows that they can begin the engineering loop. So, back to your question I didn't answer about how this fits into the relationship with universities. They leave their parents house, they go to university, and then they land a job. And if that entire experience is entirely online, that doesn't facilitate their growth.

And if they're using AI at every step, there's also a lack of of growth. So we're trying to insert specific conscious growth moments at every step. And in doing that, we're going to have to coordinate with universities to make sure that AI ethics, AI uh internals, how AI is used, how AI should be used, so that they don't externalize their brains entirely and have them atrophy just like their social skills have atrophied.

I find this across the board, this almost anthropomorphizing of AI. I remember one time I was in the car with my sister and brother-in-law and they were arguing over the instructions that came from the GPS and the GPS happened to have a woman's voice and they were saying she said this or she said that. I said stop. It's not a she. It's an it. But this relationship between the software and the individual, you know, when you get to this virtual world, it seems to be very very tight.

It seems to be very hard to break and to get and understand them. Here's humanity and here's the computer. Yeah, that's a whole podcast in itself. pre-ai, the idea that we are talking and yelling at these disembodied women in our houses, telling them to turn the lights off, whether it be Cortana or Alexa or or, you know, any of these things, it's it's it's not a healthy thing at all.

And as much as chat GPT is kind of ruining everything, at least it's not given a name. At least it's not Jeff or Anna. I don't think that's healthy. But the problem is that if a if a parrot can speak, it becomes a much more interesting pet. You know, when dogs do things like their eyebrows up or make expressions or smile, we love it. We just love that stuff.

I think when AI turns into a robot and starts walking around the house or when the Roomba starts talking back to you, you're going to see even more problematic behaviors. Uh, and the movie Her is not an aspirational story. No, it's it's not. When the Roomba tells you it should be cleaner and I'm going to go back and clean this even though you told me not to.

>> Yeah. >> You've gotten to another level. >> Then you've got problems. >> So, do you see any difference in the ability to do this preceptorship with small or midsize organizations or large organizations? Is this a problem or a challenge? I think it would certainly be easier to do it smaller because you have full control when you have two pizza teams.

Like I've always been very comfortable with 400 to 500 person companies. Those are the companies that I grew up in and worked in. That's the sweet spot where you know everybody. That was the size of my high school graduating class and that's just a good size four or 500 people. You know everybody. Even in big companies though, if you chop it up into pieces like that, like even Microsoft, as big as we are, it's really just a bunch of thousand person companies all kind of coordinating between each other.

But we've been able to do it at Microsoft uh as a pilot with 400 and found it to be successful. And I think if you do it in shards where you have executive buy off and executive support, uh it is possible. We just finished our first cohort of preceptors and it's been very successful. I just think the trick is do we have the organizational willpower to remind all of the bosses that this is much a human thing as it is a make capitalism move faster thing.

It is explicitly growing humans is the goal here. Not necessarily shipping code. And in doing that you have to acknowledge that yeah they're going to be slow and that's okay. If you take a long enough viewpoint, the actual humanity, the economic incentives should line up because happier people are more productive people are going to stay in the company.

But do you see the economic incentives working in your favor to do this? It will cost you money in the short term, but you know, it's like working out. And this is the thing like people don't work out and then they get sad that they, oh, I never worked out. Too late now. you know, not working out in your 50s is causing you problems in your 70s, too late.

Those kind of problems are going to happen when someone does not grow senior engineers and then they need senior engineers and the only place they can find them is their competitor. And they only way that they can get senior engineers is by poaching them from their competitor. And those senior engineers use their competitor's techniques and their competitor's culture and their competitor's engineering practices.

So they need to focus on growing their own people and their own culture, but they're not going to see a benefit for a couple of years. That's going to take discipline. >> Do you know of any case studies that people have written about this that are public? >> Well, we'll do our own for preceptors, but every week there's some new thing coming out from like the Harvard Business Review or somebody who is saying that AI powered engineers are getting more done, but they're lonely.

There was a really good I forgot the name of the uh I don't have it in my head right now but it was just two or three days ago they were saying how lonely people are working in AI. That's because you know the remote work plus social media plus AI has them sitting at home prompting and waiting and prompting and waiting. There was another one about flow state indicating that people are not getting into the flow state anymore and that's making them sad.

It's important for us to say the quiet part out loud. If people are sad, that's not okay and we should do something about it. I found a study, it was referenced in the Wall Street Journal, that since 2017, there's been a general decline in entrylevel hiring. Is AI part of this problem or is this something more general that is going on? I think it's more general.

I think it is short-sighted hyperoptimization in late stage capitalism. >> I think it's just very short-sighted. And by short-sighted, I mean people don't want to plan 15, 20 years out. Like people have between 20 and 40 years of productivity in them. And it seems like a long time when you're at the beginning of it, but as I'm at the end of a 35, 40year career, I'm like, "Wow, that went fast."

And I'm thinking to myself, I need to rush now to teach young people what I learned before I'm gone. Businesses need to have that kind of an attitude. Businesses need to have the same thoughtful attitude as a senior engineer who is close to retirement, but they're not. They're going quarter by quarter. So I think that it's less about AI. AI is not going to take your job.

The hyper optimization of some seuite executive is going to take your job. >> And is there any way Well, it's a loaded question, but the question comes to mind is what can we as engineers do to help? Sure, this is an ROI issue. How do we present this case to people? Well, I think speaking truth to power is important. I think saying the quiet part out loud.

Every good revolution starts at the bottom. Simply saying, I think the engineers would be happier if we did X. You know, like workers rights matter and I enjoy working with AI. I find it to be an enjoyable thing, but I find it more fun as a group as with all activities. sitting quietly and prompting is not fun. Like going from vibes to agentic software engineering techniques like I've got a really nice DevOps framework.

I can tell the AI to cut a release. It's a good release. It's it's tested. The harnesses are good. Like all the that's all software engineering stuff around it. I'm just at the very very very top now. Instead of typing build.bat, I say build. You know what I mean? It might sound silly, but I can have an idea. I can go for a walk. I can prompt remotely and I can come and see a prototype when I get back.

And then I can review the code and go through it. Like I'm looking on my other monitor here at I'm using the GitHub Copilot app, which is still has code like half the screen is still code, but it's a diff that I'm reviewing of code that was written. And then I'm commenting on individual lines saying, "Well, this is good and that's good." So, I feel like a sculptor, but I'm really the teacher in a sculpting class, and I'm walking around making small changes to everyone else's work, if that makes sense.

>> Yes. I presume a lot of this is in a context artifact of sorts. >> I'm in a session multipplexer of sorts. >> Okay. >> And I'm working through I'm trying to do a release of a thing today. And I've got five pull requests from four different people. And I need to figure out the order in which to take them. If they are architecturally correct and if they are architecturally sound, if anything's creeping in that looks like slop that is not, you know, the direction that we want to go while still maintaining a pretty

crisp clip and speed at which we're working. So like all of that is it's a very coordinate job. >> Mhm. But you're still looking at code because I remember the days, I'm dating myself now, when people used to argue over, you know, handcolded assembler as opposed to what the compiler generated. And I presume we will get to some point where the AI is writing code for other AIs and the human just don't look at it.

>> Yeah, I don't know what I think about that. Human beings have got to keep their hands on the wheel. That's the thing I will always fight for. I personally agree with you. Humans have to be in the loop. Which actually raises another thing that I've heard people talk about is that AI has killed agile. >> I don't know if you have a feeling about that.

>> I don't think it's killed it as it has taken it to its kind of ridiculous conclusion. Like everything we're doing is agile. It's just too agile. >> What do you mean by too agile? When agile in the 90s and the early 2000s was at its peak, people were calling it slop. And they were saying agile is just an excuse to go fast and break things. And when you combine agile, which is a very very quick inner loop and a very very quick outer loop, and user stories and all that kind of stuff, and then you apply AI to it, you

get wheels that spin so fast that they kind of wobble at speed and then the car potentially flies off the road. So if one believes that agile was a bad idea and they think that AI is a bad idea, combining the two just makes them super bad. But I still use conbon boards and user stories and I move them forward. So I I find that I keep coming back to agentic software engineering techniques which is agile, smart DevOps, smart build servers, smart SDLC, except we have this new power tool that has been introduced into the

mix. I'm I'm a traditionalist with a new power tool. But to get back to what we were talking about preceptorship, all this has to be taught to the potential engineer because they are not seeing this from where they sit right now. >> For example, take the SDLC. When you come out of college, do you understand the SDLC? >> So, no, you can't because you were making, you know, connect four or tic-tac-toe for your senior project.

And even a group project in college is usually three or four people. So this is why popping off the stack to the beginning of the conversation, I think that software engineering is a degree and a practice that is unique and different and discreet versus computer science. I mean I I I like my software engineering degree. I know that there are gaps in my computer science knowledge, but I think that the ideal double major would be both of those things.

But today we have to make those things part of preceptorship because that's the only way people are going to >> we do which is why I think internships cohorts senior year should almost entirely be spent in industry working and I think that that's going to require a lot of effort on the part of industry to acknowledge that we need to invest in the early and career people lest we hollow out the pipeline completely >> and then we're lost.

And that's the whole point of that paper that you mentioned. The junior engineer pipeline, the early and career pipeline will collapse if we do not agree. And then it'll and then we'll be bringing old people back out of out of retirement just like uh Y2K. And they're going to go and pull me out of retirement in 2035 to go and fix the next big date bug.

I tend to look at this as you're a farmer and you're eating your seed corn instead of planting it for the next year. The analogies are legion. >> Before we get to the architect's question here, which I like to ask all my guests, is there anything that comes to mind that we haven't talked about that you think is important? >> I keep coming back to the humanistic perspective.

Everyone's talking about the benefits of AI from a oh, it makes things better. It makes more money. It makes more this. It makes more that. But like how do we feel about this? the sociology, the psychology, like all of theologies, all of the science about are we supposed to feel this way? I think people don't sit and listen to their bodies enough. And I'll be doing my work and I'll I'll kind of go, I don't feel good right now.

Why? Why do I feel this way? Those things need to be taught. You need to go for a walk. You need to drink more water. I don't know why I feel so bad. Everything sucks. Well, are you eating well? No. Are you sleeping well? No. Are you hydrated? No. You went from sleeping all night to sitting all day to then sitting all evening to then sleeping all night.

No wonder you feel like crap. And now you're in the AI loop. You know, like we're in the matrix. We need to acknowledge what is a reasonable amount of productivity as a society. We need to understand that we cannot improve 3% every year infinitely. That is by definition cancer. To take a finite system and grow infinitely. This idea of like constant ongoing forever emergent infinite growth.

That's cancer. So if AI makes me 20% more productive, okay, 20%, maybe I get Fridays off. But that's not working. Everyone else is working triply hard right now. To what end? >> That's the problem with exponential growth. You don't realize a disaster until it's almost too late because it's like the story of the guy who asks for one corn, two corns, four corns.

>> Oh yeah. >> And and before you know it, the king has run out of food because the nature of exponential growth is you're half full the day before disaster strikes. And humans just don't have any sense how bad they are. >> Yes. >> At understanding how exponential growth works. It just doesn't. They just we can't even visualize it. >> I for one, for example, when I go to sleep at night, my cell phone is in another room.

>> That's very healthy. And this is where we need to acknowledge our our age >> and figure out how we can help the young people be better without overbiasing them and acknowledging that we are from a different generation. >> Yes. >> And we came up with different privileges and different opportunities. >> There were disadvantages but there certain advantages.

I mean we went out and did free play where we didn't have location services. The point was not to be found. >> Yep. as opposed to today where everybody wants to know where everybody is all the time. >> Yeah, absolutely true. >> One of the things that's always been interesting to me is when you define the constraint space for a project. >> Mhm. >> It's a subjective things are just as much constraints as the fact that you have latency.

I mean to pick a simple example. If you can't get the engineering skills you need, you can have the best architectural design in the world and you aren't going anywhere. So, this conversation is very meaningful to me and I hope our listeners will feel the same way. We're talking about humanizing because I like to ask several questions to my guests. What's your favorite part of being an architect?

I think it is the a thing didn't exist and then I just thunk it up and now it's a thing. That's not about architecture. That's just about being a maker, a builder. You know, my dad is a woodworker and my mom is a leather worker. And I'll just see a thing, a problem, and I'll go, "Oh, I'm going to fix that." In my garage over here, there's two buttons on the wall to open the garage.

And if I push the wrong one, it opens the wrong door, right? And it's been bugging me for 20 years. And then I got a 3D printer and I got a pair of calipers and I made a little cover on the thing. And you know, my dad's like, you could just put a piece of tape over it. >> And I'm like, well, yeah, but but I got a pair of calipers and I made a cool 3D printed colored button, you know, like a I can like it's like a nuclear button where you like flip it up and push the red button kind of thing.

Like that's way cooler. But like the point is that it didn't exist. I thought about all the 20 ways I could have solved it from cardboard to paper to painters tape to a overengineered 3D printed thing. I love being able to just make a thing that didn't exist yesterday. Another example was I wanted to get a Pelaton, but Pelatons are really expensive.

So, I went and I went to Goodwill and I got a $30 bike and then an Android tablet and then I made a website that lets me go on scenic bike rides in 3D like it's called Pedalscape. I had that idea at 10:00 and by that evening I was riding my bike through Bavaria. >> And you realize there's a whole industry now of software that enables you to do virtual rides throughout.

>> Oh yeah, I love that stuff. Yeah. But I wanted one for myself and I made it. >> What is your least favorite part of being an architect? >> A younger Scott would have said waiting, but I don't think I have a least favorite part anymore. I enjoy the waiting. You know, you can't just flash boil water to make tea. You have to kind of wait. I don't think I have a least favorite part.

It's kind of all good, honestly. I think you may have answered this before, but is there anything creatively, spiritually, or emotionally compelling about architecture or being an architect? >> I guess coming back to humans like all of this is to make something that is delightful that makes someone's life measurably better in some way. Whether it be building architecture or drawing a picture or interior design is a kind of architecture, software architecture is that thing will hold up under stress.

You know, I always come back to those high school classes where you make a bridge out of uh pickup sticks and like you know what are those things called? Ice cream sticks. >> Popsicle sticks. >> Popsicle sticks. That's extremely satisfying to like I made a thing and it also holds up under pressure. What turns you off about architecture being an architect?

>> Again, I I think that nothing but I would say documentation and vis style architecture for 10 years. And as much as I love Grady BCH, uh manually drawing UML diagrams was definitely not not my favorite part of the experience. Shout out to Grady though. >> Do you have any favorite technologies? I'm a big fan of my open- source artificial pancreas.

Like when I was diagnosed diabetic 35 years ago, it was manual shots and pricking your finger. Mine's actually it's clicking in my arm right now. I just heard it deliver insulin a moment ago and my blood sugar is as good as yours or dare I say better. Being a cyborg is a pretty favorite technology and it's a good solid architecture behind it. >> What about architecture do you love?

I think the same answer as before. Making stuff that holds up under pressure. I know that the argument of software architecture being a parallel to building architecture is a one that has been argued many many times. But you know I was in I want to in I want to say Morocco and I was looking at like the second or third largest mosque in the world. You can't be unaffected by that.

You holy crap that's a building right there. You know what I mean? >> But it's lasted >> Yeah. thousand years. You know what I mean? So, like if I could get a piece of software to last more than 20 or 30 years, I think I'd be pretty thrilled. >> What about architecture do you hate? If anything, >> I don't hate as a general rule. It's a strong feeling I try not to have.

>> What profession other than being an architect would you like to attempt? >> I am a teacher. I've always been a teacher. I'm not an architect. I'm a teacher. I've always been a teacher. Teaching is a very secret selfish thing because teaching is building, right? And I I love the feeling of being able to build minds. >> Which is probably why you realize the important of preceptorship, which other people may not have.

>> I figured it out a couple years ago that I've been teaching this whole time, so I might as well quit and teach. Well, I was an independent consultant for a long long time and I found if I taught my clients, they would be less of a burden. >> Yeah. So, teaching makes your life easier. >> Teach Amanda to fish. 100% agree. >> Do you ever see yourself not being an architect anymore?

>> Yeah, I think in 900 days when I retire, I will go and be an architect of mines. I will be a a teacher. And finally, when a project is done, what do you like to hear from the clients or your team? >> I just had the I was thinking about that movie when the very end of Babe when they go that'll do pig, you know, just just like like there's a project that I did for a company here called Pioneer Courthouse Square.

It's basically the the living room for Portland, Oregon, where it's like all these bricks in the center of the city and on the bricks are people's names and they everyone's bought their own brick. So like my grandma has a brick and you can go and find your brick. There's an app at the kiosk there where you can go and look up your name and then it'll make a print out and put an X over your brick and then you go and you find your brick.

>> I wrote that I want to say in 1996. >> Oh my. >> I'm not joking. like it's literally a 30-year-old Visual Basic app and every 5 to seven years they'll call me and make some small change, but that thing's running on Access 95 and Visual Basic and it's still working. That is very satisfying. >> Thank you very much. I found this very, very enjoyable conversation and highlighting something I think is very important for people to realize and think about. >> Yeah, it's my pleasure and nice to see you.