How It Lands

• 

  How It Lands

Inspiration

As a standup comic, I kept running into the same problem: a joke often fails not because it is bad, but because it is heard differently than intended. You usually discover that only after saying it on stage. I wanted a way to explore how a line might be interpreted before that moment.

What it does

How It Lands is a multi-agent iteration tool that shows how the same line can be interpreted in different ways. Listening-mode agents generate plausible audience reads. A reviewer agent surfaces where those interpretations diverge. The system then highlights areas worth revisiting and suggests directions to explore, without rewriting jokes or judging them.

How we built it

The project is built using Elasticsearch Agent Builder with a multi-step workflow. Agents generate interpretations, store them as structured documents, and query them using ES|QL and vector search. Elasticsearch acts as a persistent hypothesis store that allows agents to compare, filter, and reason over interpretations across runs.

Challenges we ran into

The biggest challenge was avoiding the trap of turning this into a joke scorer or laugh predictor. Another challenge was designing agents that genuinely disagreed in useful ways rather than producing shallow variations of the same feedback. Getting the multi-agent flow and storage model right took several iterations.

Accomplishments that we're proud of

We built a working end-to-end system where agents generate, review, store, and query interpretation data automatically. The result feels like a creative workbench rather than an AI critic. The system demonstrates clear agent orchestration and meaningful Elasticsearch usage.

What we learned

We learned that Elasticsearch works well as a backbone for agent reasoning when outputs are treated as data, not text. We also learned that creative tools benefit from showing uncertainty and divergence instead of hiding it.

What's next for How It Lands

Next, we want to expand beyond single lines into full scripts and allow creators to track how interpretations change across revisions. The same architecture could support other high-stakes messaging workflows where meaning matters.

Built With

• agentbuilder
• elasticsearch
• elser
• node.js
• react