AgentCraft: Gaming the AI Agent Workflow

Ido Salomon, creator of AgentCraft and MCP Apps, recently presented his innovative AI agent orchestration platform, AgentCraft, at AI Engineer Europe. The presentation, titled "AgentCraft: Putting the Orc in Agent Orchestration," offered a glimpse into a future where complex AI workflows are managed through an intuitive, game-like interface.

Who Is Ido Salomon?

Ido Salomon is the driving force behind AgentCraft, a project focused on enhancing the way humans interact with and manage AI agents. He is also the creator of MCP-UI and a co-creator and maintainer of MCP Apps. Salomon's work sits at the intersection of AI development, user experience design, and creative problem-solving, aiming to make sophisticated AI systems more accessible and manageable.

AgentCraft: A New Paradigm for AI Orchestration

Salomon opened his presentation by highlighting the magic of AI agents, stating, "Agents are magic." However, he quickly pivoted to the practical challenges of scaling these agents. The common approach of simply spawning more agents, he argued, runs into a bottleneck: human capacity to manage them. "We are the bottleneck," Salomon declared, pointing to a scenario where managing dozens or hundreds of agents requires more than just brute-force deployment.

This realization led to the development of AgentCraft, a platform that transforms the abstract concept of AI agent management into a tangible, visual experience. Drawing parallels to real-time strategy (RTS) games, AgentCraft represents AI agents as units on a map. Each agent is depicted as a character, and their tasks, files, and interactions are visualized as elements within this game-like environment.

"We need a higher ceiling," Salomon stated, explaining the need for better visibility and control over AI agent operations. AgentCraft provides this by allowing users to see what each agent is doing, which files they are accessing, and their current status. This visual feedback loop is crucial for understanding complex AI systems and for facilitating effective human-AI collaboration.

From Game Mechanics to AI Workflows

The platform's design philosophy is deeply rooted in game mechanics. Salomon explained that managing agents isn't just about commanding them, but also about understanding their progress, their dependencies, and how they interact with each other and the broader system. AgentCraft uses visual cues, such as projecting file paths as runes on the map, to make these abstract concepts concrete.

"We can actually see them working," Salomon demonstrated, showing how agents appear on the map and how their actions are represented. The system allows for direct prompting of agents, enabling users to delegate tasks, request specific actions, and receive feedback in a conversational manner. This includes functionalities like adding new features, responding to requests, and even managing collaborative efforts between multiple agents and human users.

Salomon highlighted the platform's ability to foster collaboration, not just between humans and AI, but also among different AI agents. By creating "halls" where agents and humans can convene, the system facilitates shared understanding and coordinated action. "We're also collaborating with the agents," he noted, pointing to a chat interface within the platform where agents communicate their status and humans can provide direction.

The Future of Agent Collaboration

The presentation showcased how AgentCraft allows users to see not only what their agents are doing but also what other agents are working on. This transparency is key to avoiding conflicts and optimizing workflows. The system also includes features like "hand-off" mechanisms, allowing agents to transfer tasks or responsibilities to other agents or humans, creating a more dynamic and efficient collaborative environment.

Salomon concluded by emphasizing that the skills developed through gaming – strategic thinking, resource management, and understanding complex systems – are directly applicable to managing AI agents. "We've been preparing for this our whole lives," he stated, suggesting that the intuitive, visual approach of AgentCraft taps into innate human capabilities. The goal is to move beyond simply coding AI to a future where humans and AI collaborate seamlessly, with agents acting as extensions of human intent, supported by robust visibility and communication tools. Salomon invited the audience to download and experiment with AgentCraft, highlighting its experimental nature and the ongoing development to further enhance its capabilities.