So What About AI Agents Agentic Prediction - EP 52 - Michael Ulin - Tenki AI
Jan 27, 2026
Michael Ulin, CEO and co-founder of Tenki AI and serial AI founder, explains building multi-agent forecasting systems. He covers agent architectures, transparent limitations to build trust, detailed logging and evaluation, avoiding overfitting via rapid feedback, and practical launch strategies like validating demand and bootstrapping.
AI Snips
Chapters
Transcript
Episode notes
Decompose Forecasts For Better Accuracy
- Break forecasts into subcomponents and assign agents to each task.
- Use debate-style agents to challenge base assumptions and improve accuracy.
Log Everything And Run Controlled Tests
- Log every input, agent output, and intermediate step to enable debugging.
- Use recorded evaluations to run controlled experiments and improve prompts.
Use Fast-Resolving Markets To Avoid Overfitting
- Use prediction markets for rapid feedback to avoid overfitting and speed iteration.
- Validate models on many daily-resolving events rather than rare, slow outcomes.