Earlier this week, we launched a new episode of our “Tech Connects” podcast with guest Adam Seligman, who’s VP of developer experience at Amazon Web Services, or AWS. We talked about how AWS is launching a portfolio of A.I.-powered helpers and tools for developers, and how the tech industry is changing rapidly thanks to the evolution of A.I. and other technologies.
Here’s the video from our discussion:
Here’s the main takeaway from our talk: Generative AI will change the landscape of software development, but it won’t curb demand for tech talent. There will still be a need for highly skilled, specialized developers, but generative AI will make it possible for people with less experience to enter the field and contribute to generative AI and other kinds of tech teams. This is because generative AI tools will allow people to experiment and learn new technologies more easily; this could lead to a more democratized approach to software development, where more people can participate in creating software.
Here are some excerpts where we dig into those issues; check out the full episode for more!
Q: One of the things that was interesting was this idea that you were throwing out there about how when you talk to most folks about generative AI and machine learning and so on, there's quite a bit to focus on. There’s the upper echelon of talent, like the super highly specialized PhDs who are busy every day with training, perfecting LLMs and then sort of figuring out how to integrate that into workflows and products and all the rest of that good stuff. But what you were suggesting is that more junior people on a generative AI team—in terms of people who may have taken a few AI classes and so on—are just as valuable for these teams in terms of training and perfecting LLMs. That was fascinating to me. I just wanted to kind of dig into that a little bit like what you mean by that.
Adam Seligman: I think generative AI capabilities are going to flip a bunch of assumptions about being a software developer, working as a software developer, or even becoming a software developer. It’s going to flip those assumptions on their head. As an aspiring developer you can discover new technologies, ask for a little help, get something explained to you in a way that makes the most sense to you try it, instantly, like you never could before. Those kinds of capabilities at the fingertips of anyone that wants to play with a cloud or play with technology is extraordinary.
It's really fascinating to me because, before the whole LLM AI revolution kind of truly got underway, there was still a push for no- and low-code tools, and you saw everyone from startups to major enterprise companies trying to democratize (for want of a better word) the ability of people to sit down and, you know, they might have had some coding knowledge but if not, and even if they didn't have an extensive knowledge, they could still produce a kind of a simple app. It seems that this AI revolution is going to accelerate that even further, but when I talk to different people, it seems like there's different timelines about it. You talk to some people, and they say, “Knowledge of the fundamentals of coding is still going to be essential, and an era where you can just sort of issue a prompt and get a piece of complete awesome software back is a long way—if ever—from coming to us.” But then you talk to other people who are more optimistic where they're saying, “Oh, developers are going to be able to spin up software instantly within a couple of years.” I'm just wondering where you fall in that [scope], because you're in a privileged position to see where things are evolving.
Adam Seligman: We're all learning every day, but with our customers and our community we see generative AI innovation happening everywhere. We see our customers are the most skilled data scientists and ML engineers, building and training custom models with Sagemaker on their own corporate data. We see skilled developers building complex systems using Bedrock and applying AI to build things like agents and intelligent automation in the backends. We see developers of every skill level including early career interns skilling up incredibly fast because they have generative AI tools like our queue developer product and there are others in the market. Also, lots of choice because everybody gets new capabilities that help them take where they are and just push the envelope forward.
Q: the evolution has been astoundingly rapid. It seems that every time I log in to any sort of generative AI tool and try it, there's all these like new features that have been added and so on, and it's astounding.
Adam Seligman: We were talking before about a moment where generative AI will just write a giant output of code and replace coding for you, and I don't think that's exactly the right way to think about it. The thing we're seeing with our customers and our developer community is everybody can move really fast all at once right now. They can ask for approaches. They can get help migrating old Java code like with our queue developer transformation feature, our code transformation feature. They can ask for a serverless architecture, how to build an event-driven pattern to solve a certain kind of problem. Wherever they are in their journey and the kind of problem they're solving, they can get help right away and so to have that kind of help again available for you right now, just in time, is extraordinary in how it will enable humans to do more.
Q: And then the tools you're working with there. There's an element baked into it as well, right? So, when you're asking for say a code snippet or something just to be able to sort of verify that what you're getting back is—I'm trying to think of the right word here… relatively…
Adam Seligman: Yeah, well, no software's ever done, right? We all know that we're on this journey to build things, then add more features and fix problems and learn how to operate at scale and run them in more places and add new features. So, no software is ever done. You must sort of bring that mentality. We believe the best way to equip developers is not just with one modality like in-line code completion or chat but really help them across the entire software development lifecycle. It might be learning, exploring solution architecture. Planning an approach to build an application scaffolding code for you like in-line, possibly writing tests for you, suggesting improvements in your code, which might be like a chat modality migrating code. Like from Java 8 to Java 17, our Q product helps with that. Debugging operational issues. So, you might have some challenge in getting an error message you're not familiar with. “Help me understand,” that so it might be a chat modality, but all the way in the operations side, not in the software development side. At Amazon, we're trying to help customers and our developers with every fast of the entire software development lifecycle and it looks like it's going to create an extraordinary up-leveling of every developer and their capabilities.
Q: When you think about companies bringing on junior developers, obviously they go through the whole interviewing process where they're assessed. But even if they have tons of skills already, some experience, whatever… we talk to hiring managers and project managers and so on, and there's always a concern about: will people be able to upgrade their skills rapidly? Will they be able to introduce themselves into the workflow that's currently in process and then gear up and be active robust contributors? And it seems like AI would really help with that. It seems that that's something where you'd be able to kind of catch up on the flow more quickly than before, and that's I mean that certainly strikes me as awesome potential benefit.
Let's say you’re part of a developer team where you're tasked with something like building models or training models or something like that. A lot of people who are coming onto these teams, especially interns and more junior developers, might be great prompt engineers. They might be awesome at sort of querying. But in terms of making that leap to the architecture or the training side of it, what skills do they need to have to like kind of fully operate quickly on AI teams?
Adam Seligman: I think it's easy for those of us that have been in this field a while to have a preconceived notion that you have less capabilities when you're earlier in your career and as you go, you get more capabilities and get super deep in some areas. And what I think is fascinating is with the power of the cloud and generative AI, you can ramp to be incredibly valuable at something incredibly deep without necessarily having all those years of [training]. I give you an example: Sagemaker allows you to train or fine-tune your own models incredibly easily and you don't have to know Assembly language, you don't have to know C. It’s pretty extraordinary what you can do. So, you're if you're an enterprise company or big customer, your interns may be training your foundation models in Bedrock or Sagemaker; your interns may not just be doing prompt engineering but doing the quality engineering that ensures these generative AI systems are high quality. So, lots of preconceived notions are changing in this world of generative AI, and I am an optimist can find it incredibly exciting that early career people can now suddenly have superpowers and provide a way outsized impact of benefit to their company, building great software and building great solutions and solving problems.
