​Featured Tracks:

1. ∞ REAL Agent Challenge:
The ultimate test for autonomous agents. Compete on the REAL global leaderboard to build the most reliable, efficient, and versatile browser agent.

• Subset benchmark with ~16 tasks: Small, curated set pulled from REAL-style tasks. Your agent must solve these reliably under real constraints
• Speed is weighted equal to accuracy: Latency directly affects score. Large slow models will tank; efficient pipelines win.
• Hidden test set for final eval: You can develop on the public 16 tasks, but scoring uses unseen variants, so hard-coding won’t work.

2. Vertical Agent Track
Rapidly develop a production-ready vertical agent using powerful agentic infrastructure, all in a 2-day sprint.

• Leverage the AGI API: Utilize robust, agent-centric components (advanced memory, tool orchestration) for reliable, focused solutions.

• Vertical Focus: Solve a specific, high-value domain problem (e-commerce, data analysis, recruitment) with a targeted agent.

• ​Lovable Credits: Build a high-quality demo-ready front-end using Lovable and integrate voice with Telnyx.

• Demo Opportunity: Showcase your solution to industry experts for recognition and potential investment.