Agentic Development

GitHub Next develops Chopin for visual, real-time agent planning

GitHub Next is building Chopin, a tool that uses visual artifacts and real-time chat to structure plans between humans and agents. The project aims to address the limitations of text-based interfaces and improve how junior developers learn from senior engineers' agent interactions.

· 1 source · Human-agent collaboration

GitHub Next, the R&D branch of GitHub, is developing a project called Chopin to facilitate collaborative planning between humans and agents. The project aims to move beyond text-heavy CLI interfaces by using visual, interactive artifacts and real-time chat to structure plans. In the current prototype, free-form chat is classified by background agents into questions, proposals, opinions, and constraints, which are then organized into structured plan artifacts. The team is exploring the integration of tldraw whiteboards and real-time transcripts to allow agents to observe human conversations and proactively insert research reports into the plan. The goal is to create boundary objects that bridge the gap between human visual and spatial reasoning and agent language processing, allowing teams to align on architecture before handing off to implementation agents. Maggie Appleton, of GitHub Next, argues that current text-based interfaces prefer agent preferences over human cognitive needs, and that tool builders must create social, real-time, and visual environments to effectively collaborate with agents. Appleton notes that current AI models are poorly suited for UI design and CSS implementation because they are trained primarily on language and code rather than visual inputs. She predicts that general models will achieve baseline aesthetic usability within one to two years as labs increase multimodal training, though she notes that aesthetic standards are historical and cultural, making them difficult to bake into static model weights. The shift to agentic development is also altering the learning curve for junior developers by making senior engineers' planning processes visible. In traditional pair programming, juniors learned by observing seniors; in the agent era, this is replaced by observing how seniors prompt agents, ask questions, and correct outputs. GitHub Next prototypes are designed with open-by-default agent chats within teams, allowing all members to see each other's interactions. This visibility helps juniors understand the quality of plans and the specific pitfalls that seniors address. Appleton suggests that agents can also act as proactive guides for juniors, asking more critical questions or providing context that a junior might miss, effectively bridging the experience gap without requiring juniors to have full senior-level intuition.