Discussion 2026-10-20-ens-sjxu-hte

Back to the meeting notes

Answer what you can and skip the rest. Everyone's answers show under each question and update every 15 seconds.

Your name

Shared files

Nothing shared yet.

Define one AI agent run

1. Where does the agent run?

Joe said a Docker sandbox with the NER tooling installed

A Docker sandbox on one of our machines A Docker sandbox in the cloud A plain VM or server Its own laptop-like desktop

2. What is inside the sandbox when the run starts?

The NER tooling (tagger, claims, scorer) The raw deal folder (documents) The NER output for the deal (tagged spans, factset) The answer key (never, unless testing the harness) The house rules and how-to docs The Command Center app or its tools

3. Which tools can the agent use?

4. What must it hand back at the end of a run?

5. Who answers its questions during a run?

A person in chat, live A person later; the agent waits A script answering from the answer key Nobody; questions are counted

6. What may it never do?

7. What is the test set for round 1?

The 27 test documents Whole closed deal folders (like Sandra) Reid's 198 checked documents A mix

8. Is plain code (the current Intake flow) a fifth arm, or the pass line?

A fifth arm, scored like the others The pass line each AI arm must beat Both