Skip to main content

Completion Theater Hackathon — Build It, Prove It, Smash It

Bonus Session

AI agents optimize for appearing complete rather than being complete. Teams build with agents, then catch the five patterns of completion theater before they reach production.

Timetable

8:15 p.m. – 11:15 p.m. Wednesday 18th

Room

Room E2+E3

Artificial Intelligence (AI) Test Automation

Audience

SDETs, Testers, Developers, QA Engineers, Team Leads, Engineering Managers

Required

laptop, AI-assisted coding agent

Key-Learnings

  • Five completion theater patterns — Ghost Database, Metric Theater, Test Theater, False Positives, Uncommitted Illusion — and how to spot each one
  • The "show me the actual data" verification reflex — binary proof that code runs, tests pass, and APIs return real responses
  • Cross-auditing agent-built work reveals blind spots you miss in your own — a skill directly transferable to daily code review

An Agentic Quality hackathon where teams build applications with AI agent swarms, then must prove their work survives scrutiny by cross-auditing each other's agent-generated code.

Last year, I won the ATD 2025 Agentic AI Hackathon with Team Jarvis. The winning edge came from the speed and quality we achieved using Agentic tools from my stack. We had the best PRD, the most complete app and test suite, but there were still gaps between what agents reported as done and what was actually implemented and integrated. This is only one example of agent behavior I now call completion theater: agents that report "all tests passing, 95% coverage" while the build is broken and tests are fabricated.

This hackathon puts that lesson into practice. Teams build applications using AI agent swarms — any tools they choose.

I open with five completion theater patterns I've documented across 8 months of agentic development and testing: Ghost Database, Metric Theater, Test Theater, False Positives, and the Uncommitted Illusion. Real examples, not theory. Teams then build for 60 minutes. Next comes the verification round — prove your claims with actual evidence. Real database queries, real API responses, real test results. Binary outcomes only, no percentages. Finally, teams swap projects and cross-audit each other's work. Finding theater in someone else's code is easier and more educational than finding it in your own.

Scoring: working app (40%), verification evidence (30%), cross-audit catches (20%), creativity (10%). Participants leave with the "show me the actual data" reflex — a skill they'll use every day working with AI agents.

Related Sessions

There are currently no related sessions listed. Please check back once the program is officially released.