Inspiration
We wanted to explore what the web looks like when AI agents are treated as first-class users.
Robotics procurement felt like the perfect use case. Choosing cameras, LiDAR, compute, sensors, and power systems involves a lot of searching, comparing, budgeting, and repetitive decision-making.
Our idea was simple: let the AI handle the research and recommendations, while keeping the human in control of the final decision.
That became Human-Agent Robotics Procurement
What it does
Human-Agent Robotics Procurement lets an AI agent work directly with a robotics procurement workspace through WebMCP.
The agent can search equipment, compare products, calculate budgets, and generate complete configurations. Instead of simply returning a list of products, it can reason about the project and propose different configurations based on priorities such as cost or performance.
The human remains in control. Recommendations are presented for review, and consequential project changes require explicit human approval.
The agent proposes. The human decides.
How we built it
We built the application with React Native, Expo, TypeScript, and Zustand, with WebMCP providing the agent-facing interface.
We exposed six structured tools through document.modelContext for product search, comparison, budgeting, project updates, optimization, and human approval.
The application maintains shared project, activity, and approval state, allowing the human UI and AI agent to work with the same underlying workspace.
A key part of the implementation was separating recommendation from mutation. The optimization agent can generate configurations without changing the project. Only an explicit human approval can turn a recommendation into an actual project change
Challenges we ran into
The biggest challenge was figuring out how to integrate WebMCP into an existing Expo application while keeping browser-specific functionality separate from the native application.
We also had to carefully design the boundary between agent autonomy and human control.
It was tempting to let the agent automatically apply its recommendations, but that would remove an important part of the procurement decision. We instead designed a workflow where the agent can do the time-consuming work independently, while the human remains the final authority.
Making the agent's actions visible was another challenge, so we added activity tracking to show users what the agent is doing and when human input is required
Accomplishments that we're proud of
We're proud that we transformed a normal procurement interface into an agent-ready web application.
The agent can discover and use structured capabilities directly from the page instead of trying to understand the application by interpreting the UI.
We successfully implemented six WebMCP tools and connected them to the same application state used by the human interface.
We're especially proud of the approval workflow. The agent can search, analyze, optimize, and recommend independently, but it cannot silently make consequential changes.
That creates a practical model for human-agent collaboration:
AI handles the work. Humans retain the authority
What we learned
We learned that building for AI agents requires thinking beyond traditional UI design.
A human sees buttons, screens, and menus. An agent needs clearly defined capabilities, inputs, outputs, and consequences.
WebMCP gave us a way to expose those capabilities directly to the agent.
We also learned that meaningful agent autonomy doesn't require removing humans from the process. In many real-world workflows, the best experience may be an agent that can do most of the research and reasoning while keeping humans responsible for important decisions
What's next for human-agent-collab
We're excited to take the project from a prototype into a more complete autonomous procurement system.
Next, we want to connect the workspace to real supplier catalogs, live pricing, inventory, quotations, and delivery information. We also want to introduce multiple specialized agents that can handle supplier discovery, technical compatibility, cost optimization, and negotiation.
The long-term vision is a procurement workspace where AI agents can take on increasingly complex work across the entire purchasing process, while humans remain in control of approvals and high-impact decisions.
Human-Agent-Collab is our exploration of what procurement could look like when humans and AI agents work in the same workspace not as separate systems, but as collaborators
Log in or sign up for Devpost to join the conversation.