Aditya Challapally
This as-told-to essay is based on a conversation with Aditya Challapally, a 30-year-old Microsoft employee who teaches a course for Stanford Online about generative AI. This story has been edited for length and clarity.
I started working in AI about a decade ago. I started as a data science intern at Uber, then did AI consulting at McKinsey, and later joined Microsoft, where I now work on Copilot.
I started guest teaching at Stanford four years ago and recently co-created a course called Mastering Generative AI for Product Innovation, which launched on Stanford Online in August 2024. It's an online, self-paced course that runs throughout the year. All of the research comes from talking to 300-plus users and 50-plus executives.
A lot of the people who take the class are tech adjacent, such as customer support representatives for a technical product, or product managers for a software or hardware product. They'll often be working on somewhat of a technical product and the course helps them understand gen AI a little bit more.
We teach three modules in this course. The first module explains what Gen AI is and where the biggest opportunities are. In the second module, we talk about what great Gen AI products look like.
The third module talks about how great Gen AI products are built and what individuals can do to set themselves up to be more influential, relevant, and useful when building Gen AI products.
These are the two main pathways you can take to do so.
When I go out and talk to Fortune 500 leaders, they say that their most burning need is for professionals who bridge both worlds — those who understand the business requirements but also understand the technical requirements.
This doesn't necessarily mean that you have to learn how to code, but you at least need to have enough technical literacy that you can translate product visions into technical requirements.
The beginner version is just getting really good at prompt engineering. This sounds like it would be quite basic, but understanding the exact limitations of prompts and all of the different tools across text, audio, and image makes you already very valuable in a business setting because you can help generate ideas even before they get to the technical team.
At an intermediate stage you also should start to understand a little bit about how gen AI systems work in systems design, like how gen AI models can be called within your data boundary.
Companies have data boundaries for which they have an agreement with their customers that their data can't go beyond. So if you're a bank, you may have an agreement with your customers that only the bank will use their information. If you send that in some sort of chat to OpenAI, that would be breaking the company data boundary. So something as simple as knowing that is already really helpful.
In the advanced stage of this track, there are two options.
Some people who don't work in big companies go deeper into understanding coding a little more. People who work in Big Tech companies usually dive deeper into system architecture. So they'll understand things like data boundaries and data flow diagrams in a lot more detail.
The domain expertise track is where business people automatically lean toward and have an advantage. This is not necessarily knowing more about the industry, but knowing how gen AI can apply to the domain in more detail.
For example, in finance, you have to know things like what data you can use to train a specific model. You also have to know things like what types of privacy and security regulations you have to go through to get an app approved or release a gen AI-related app.
This skillset is so valuable that companies pay large amounts to consultants that have this specialized expertise. I know this guy who used to work as an operations manager at a bank and he figured out where gen AI was the most valuable. Now, companies will just call him to figure out where to launch their gen AI product.
The best thing I see people do is try to automate a lot of their lives with gen AI. They use ChatGPT or Claude for everything and that helps them understand the limitations of AI really well and how to prompt it.
When beginners start to use gen AI, they're not used to what I call the abundance of intelligence. They'll say "Can you give me a response to this text message?"
Experts who use gen AI a lot will say something like, "Can you give me 20 responses to this text message?" And then they'll go and use their taste to pick one.
Outside of work, I use it in many ways to think through a lot of plans. It's really helpful as a thought partner for me, even if for communication, for general planning, or for something even as banal as trip planning.
Instead of asking a friend for advice you should think about asking an LLM or a chatbot for advice. That's when you really start to understand how it's useful.
