After you submit a research request, whether it is a competitive landscape for a therapeutic area or a deep dive on a specific asset, the AI Research Agent returns a structured response with transparent reasoning, downloadable deliverables, and full source traceability. This article covers what you see when a task completes and how to work with the results.

The Agent supports two execution modes:

Plan Mode shows you the research plan before execution. You can review, edit individual steps, adjust deliverables, and approve before the Agent begins. Use this when you want control over scope and output structure.

Auto Mode lets the Agent determine the steps and deliverables on its own and execute without requiring plan approval. Use this for straightforward queries where speed matters more than step-by-step control.

If you start in Chat mode and your question would benefit from a structured workflow, Chat will suggest switching to Agent mode. You can continue in Chat or switch to Agent to get structured deliverables.

When a task completes, the response includes four sections:

A list of the steps the Agent took to fulfill your request, showing how it interpreted your query and arrived at the final output.

A synthesis of findings and conclusions from the research. For queries that do not require deliverables, this summary may be the primary output. The summary is downloadable in Word format by clicking its icon in the Files Created panel.

If your plan included deliverables, they appear alongside the analysis summary:

Reports, Smart Tables, and charts open directly within Maven Bio. Other file types prompt a download.

All outputs for a task are listed in the Files Created panel on the right side. This includes deliverables and Research Notes, which capture supplementary observations, research gaps, or deeper exploratory analysis the Agent performed while working on your query.

All Agent outputs include inline citations. As you scroll through the analysis summary, citation links appear in the side panel. Click any citation to go directly to the referenced source. This applies across all deliverable types, not just the summary.

When the Agent completes a multi-step workflow (for example, a competitive landscape analysis for an oncology indication), you can save the planning logic as an Agent Blueprint. The next time you need the same analysis for a different indication, drug class, or company set, open the Blueprints panel in Agent mode and select your saved workflow. The Agent applies the same research steps to the new subject.

Agent Blueprints store the Agent's planning logic and step sequence. They are separate from Smart Table and Report Builder blueprints in your workspace. To reference a Smart Table or Report blueprint inside Agent mode, use @blueprint in your message.

After reviewing results, you can ask follow-up questions within the same Agent session. The Agent retains context from the completed task, so you can refine the analysis, request additional deliverables, or pivot to related questions without restarting.