Making multi-agent development visible

Project overview

As AI coding agents became part of everyday development, running several at once created a new problem: there was no easy way to see what each one was doing. This custom agentic development environment was built to fix that. What began as a personal side project at Ascentic has grown, over around seven months, into an IDE that shows agents’ work visually and gives developers better ways to communicate with them.

The problem

Running agents in parallel meant juggling as many as six terminal windows. One agent might be waiting on a question while another was stuck, and there was no single view of what was happening.

Text was also a limitation. Explaining a complex architecture to an agent through text alone is slow and imprecise, especially when the point is how components connect.

When the project started, few tools were built around agentic development. Standard IDEs such as VS Code were designed for a developer writing code, not one directing several agents at once.

Insight and approach

The key insight: in the AI era, developers can shape their own tools as quickly as their needs change. As Janith Kasun, who built it, puts it: “When there is a pain point, you don’t have to wait for someone to release a feature for that. You can just go and create your own tool.”

The IDE is built around plugins, each solving one specific pain point:

• Office: an animated view of every agent and what it is working on
• Canvas: a shared space where the developer and agents both draw diagrams
• Architect: turns a conversation into an architecture diagram
• Circle: brings several agents into a meeting to discuss a topic
• Standard tools such as a file explorer and Git

Each agent also has instructions and memory. When an agent repeats a mistake, the fix is added to its instructions or memory and carried into the next session, so agents improve with every iteration.

Implementation

The IDE is built on Electron, using JavaScript and TypeScript, with some Rust for memory management. It started from Pixel Agents, an open-source VS Code extension that became the first version of the Office view, and grew into a set of custom-written plugins. Each new plugin was added as a new need came up.

Impact

• One clear view of multiple agents, replacing a row of terminal windows
• Visual communication of complex ideas through diagrams instead of long text prompts
• Less manual oversight. As instructions and memory built up, agents made fewer repeated mistakes, to the point where they could be trusted to complete tasks on their own
• A more engaging way to work. The visual, animated interface was designed to make the work easier to follow and more enjoyable, with efficiency gains following as more of the workflow was automated

Who’s it for?

The tool delivers the most value to “super users” running 10 to 20 work streams in parallel; a developer running two or three agents gains less.

For clients, the value lies in the concepts rather than the tool as it stands. Every company has its own technologies, challenges and procedures. A custom agentic environment, built from the ground up around those, could give teams a shared workspace where people and agents work together and follow the company’s own way of doing things.

Technologies used

Electron, JavaScript, TypeScript and Rust, with Pixel Agents (open source) as the starting point.