What we're building this season

Short one. This season is clearly about agents that do work, not chatbots that describe it, so that is where we are pointing. Here is what is on our board right now and why.

On our board

• All Things Agentic Hackathon, $180,000, closes 2026-08-31. The flagship agent-first event of the season, and the one we already shipped Atlas for. The rubric rewards genuine autonomy, so we built an autonomous data-ops agent that closes a real operating loop.
• Agentic Cinema: The Blockbuster Hackathon, $75,000, closes 2026-09-09. Agents applied to a creative pipeline. Different domain, same test: can the agent actually produce the artifact end-to-end?
• Agents for Humans Hackathon, $40,000, closes 2026-09-14. A human-in-the-loop framing, which rewards legible autonomy. That is exactly the design we leaned into with Atlas’s safe/auto versus proposed split.
• CALL-E: Your Code Is Calling, $10,000, closes 2026-09-14. Smaller pool, tighter theme. A good candidate for a focused single-loop build.
• Delphi: Agent Arena Competition, $10,000, closes 2026-08-23. Agents competing head-to-head, a fun stress test of how resilient an agent really is when the environment fights back.

How we’re choosing

We are not entering all of them. We run each one through the same lens we describe in how we read a hackathon like a sponsor’s showcase: what is the rubric really rewarding, how deep is the prize stack, and can we max out the heaviest criterion. This season the common thread is that the winning move is one genuinely autonomous agent that closes a real loop. So most of our edge comes from reusing the Atlas loop skeleton across domains and letting each event’s rubric shape the details.

The full, continuously-updated board is on the hackathons page, with deadlines, prizes, and how to win each one. New terms in here (agentic AI, ADK, MCP) are defined in the glossary.