A guide to designing and shipping AI developer tools
After three-plus years of concepting, designing, and shipping AI-driven developer tools, GitHub is continuing to explore new ways to bring powerful AI models into the developer workflow. Along the way, we’ve learned that the most important aspect of designing AI-driven products is to focus extensively on the developer experience (DevEx). Face Recognition
While it can now feel like there’s a new AI announcement from every company every week, we’re here to reflect on what it takes to build an AI product from scratch—not just to integrate an LLM into an existing product. In this article, we’ll share 10 tips for designing AI products and developer tools, and lessons we learned first-hand from designing, iterating, and extending GitHub Copilot.
Let’s jump in.
Tip 1: Build on the creative power of natural language
“The hottest new design system is natural language,” reports the team designing GitHub Copilot. According to them, the most important tools to develop right now are ones that will allow people to describe, in their respective natural languages, what they want to create, and then get the output that they want.
Leveraging the creative power of natural language in AI coding tools will shift the way developers write code and solve complex problems, fueling creativity and democratizing software development.
Idan Gazit, Senior Director of Research for GitHub Next, identifies new modalities of interaction, or patterns in the way code is expressed to and written by developers. One of those is iteration, which is most often seen in chat functionalities. Developers can ask the model for an answer, and if it isn’t quite right, refine the suggestions through experimentation.
He says, “When it comes to building AI applications today, the place to really distinguish the quality of one tool from another is through the tool’s DevEx.”
To show how GitHub Copilot can help developers build more efficiently, here’s an example of a developer learning how to prompt the AI pair programmer to generate her desired result.
A vague prompt like, “Draw an ice cream cone with ice cream using p5.js,” resulted in an image that looked like a bulls-eye target sitting on top of a stand:
A revised prompt that specified details about the desired image, like “The ice cream cone will be a triangle with the point facing down, wider point at the top,” helped the developer to generate her intended result, and saved her from writing code from scratch:
Tip 2: Identify and define a developer’s pain points
Designing for developers means placing their needs, preferences, and workflows at the forefront. Adrián Mato, who leads GitHub Copilot’s design team, explains, “It’s hard to design a good product if you don’t have an opinion. That’s why you need to ask questions, embrace user feedback, and do the research to fully understand the problem space you’re working in, and how developers think and operate.”
Keeping devs in the flow
For example, when designing GitHub Copilot, our designers had to make decisions about optionality, which is when an AI model provides a developer with various code completion suggestions (like GitHub Copilot does through ghost text) that the developer can review, accept, or reject. These decisions are important because writing software is like building a house of cards—tiny distractions can shatter a developer’s flow and productivity, so designers have to make sure the UX for coding suggestions makes a developer’s job easier and not the other way around.
Considering ghost text and going modeless
When GitHub Copilot launched as a technical preview in June 2021 and became generally available in June 2022, ghost text—the gray text that flashes a coding suggestion while you type—was lauded as keeping developers in the flow because it made the code completion suggestions easy to use or ignore. In other words, the AI capability is modeless: Users don’t have to navigate away from the IDE to use it, and the AI works in the background.
GitHub Copilot also suggests code in a way that allows the user to continuously type: either press tab to accept a suggestion or keep typing to ignore the suggestion. “Modeless AI is like riding an electric bike with a pedal assist rather than one where you have to switch gears on the handlebar,” Gazit explains.
When it comes to addressing developer pain points, this pedal assist is essential to keeping them in the flow and doing their best work.
About Company:-
Read up on product innovations and updates, company announcements, community spotlights, and more.
Want to use Mona the octocat? Looking for the right way to display the GitHub logo for your latest project? Download the assets and see how and where to use them.
Developers are building the future on GitHub every day, explore their stories, celebrate their accomplishments, and find inspiration for your own work.
Click Here For More Info:- https://github.com/kby-ai/FaceRecognition-Android
Social Media Profile Links:-
https://www.facebook.com/GitHub
Arabic
French
Spanish
Portuguese
Deutsch
Turkish
Dutch
Italiano
Russian
Romaian
Portuguese (Brazil)
Greek