◆ Assistive Specialist
Models the sell-side question set from the quarter, the guidance and each analyst's prior lines of probing, and drafts framing for each. Strictly advisory: the IR oversight agent decides what to say, the disclosure judge clears it, and the board owns apex external accountability.
Memory
Working The draft Q&A set and the quarter's key messages.
Episodic Prior calls' questions and how each analyst tends to probe.
Semantic Disclosure boundaries, guidance language, competitor positioning.
Store Transcript corpus + analyst-coverage knowledge graph
Orchestration
MCPA2A
Harness · Managed Agents … drafting session with disclosure guardrails.
Tools
⌕ Earnings call transcript corpus Retrieval { } Analyst coverage + estimates feed API ⇄ Disclosure-compliance judge A2A ⇄ IR oversight agent A2A
Evals & guardrails
- Disclosure judge flags any suggested answer that risks selective disclosure.
- Strictly advisory … the IR oversight agent decides and delivers; this agent never speaks externally.
- Coverage check: did the drafted set anticipate the questions actually asked?
Offline reflection
Replays the actual call Q&A against its predictions to improve question anticipation each quarter.
Frontier edge
- ▲World-model simulation: war-games the call as an adversarial Q&A, modelling how each analyst is likely to push and where a follow-up could corner the IR oversight agent.
- ▲Eval-gated continual learning: every quarter's actual call refines its per-analyst question model, with a disclosure-boundary check before any new framing is adopted.
- ▲Formal action-gating: drafted answers carry a signed disclosure-boundary attestation, so anything risking selective disclosure is provably blocked before it reaches the IR oversight agent.
In numbers
under 1 hour
Median Q&A book build
tracked per call
Anticipated-question hit rate
Handoffs
Fed by ← Earnings Prep Agent