The age of AI is here and software developers, who are already adopting it at scale, will be one of the biggest beneficiaries of this productivity-multiplying technology, GitHub’s CEO Thomas Dohmke told us during his recent visit to India.
But Thomas said that to make the most out of AI, developers need to upskill in the new field of prompt engineering because just asking an AI model a question doesn’t guarantee that you get exactly the answer you are looking for.
Instead, Thomas says, you should first get used to the process of refining your queries. “If you tell an AI model to write a letter or a resume, it will always give you something. But, the next step is to refine your query.”
AI models, he says, are probability machines that simply predict the next word or code. “So, we have to understand how to use that machine to get the right output.”
For coders in complex business environments in particular, AI models might need additional contexts to give the desired code output. “You might need to feed the AI model specific files, for example,” Thomas says.
On the contentious issue of whether students should even bother learning programming languages anymore in a world where AI systems are getting more and more proficient in them, Thomas unequivocally says, “keep learning those languages”.
“Human language is not deterministic. Two people can say the same sentence and they can mean very different things in almost all human languages. Code that instructs the machine, it’s an abstraction layer on top of the machine, is deterministic. The same line of code will do the same thing every single time. And so, you have to have both of these tools in your toolbox.”
Human language in combination with an AI code generation platform, like GitHub’s Co-pilot, enables developers to write a lot of code today by just summarising what they want to do, says Thomas. “We announced a product called Co-pilot Workspace in April that lets you do exactly that.”
Coding with Co-pilot
Co-pilot Workspace comprises three AI agents. The first agent helps refine your project idea. “It helps you reflect on your idea and makes it more specific than a product man ager typically would be able to do. Often when you make feature requests on an app, it’s not specific enough for the developer to actually implement that, and so they go back and forth (with the person who gave the request) to figure it out. So, that’s the first agent.”
The second agent looks at what files you need and figures out how you can implement your task. “That’s challenging. I’ve been a software engineer for years, but if I were to join a software company in Bengaluru, the first thing I would have to figure out is how they operate. And then I have to figure out how things work, what’s in which file, what did the engineer think when they built a particular thing, and how I can now extend it.”
The third agent takes those instructions and implements them. “And along the way, you interact with the model to tell it what it got right and what you want to change.”
At the end of this whole process, Thomas says, you have code that was written with the help of AI.
Originally Appeared Here
